From 4febf3b9613eff30ad4bcb4ad81c3594bbab7b44 Mon Sep 17 00:00:00 2001
From: arman <arman.davtyan@xfel.eu>
Date: Wed, 7 Dec 2022 09:25:51 +0100
Subject: [PATCH] first commit
---
README.md | 11 +
data_drift_check.py | 34 +
inv_eval.py | 1627 ++++++++++++++++++
inv_inference.py | 69 +
inv_train.py | 103 ++
src/data/__init__.py | 0
src/data/data.py | 326 ++++
src/data/data_preproc.py | 99 ++
src/models/find_components/Find_Component.py | 194 +++
src/models/find_components/__init__.py | 0
src/models/fit_methods/Fit_Methods.py | 302 ++++
src/models/fit_methods/__init__.py | 0
src/models/fit_methods/model.py | 164 ++
src/utils/__init__.py | 0
src/utils/utils.py | 238 +++
15 files changed, 3167 insertions(+)
create mode 100644 data_drift_check.py
create mode 100644 inv_eval.py
create mode 100644 inv_inference.py
create mode 100644 inv_train.py
create mode 100644 src/data/__init__.py
create mode 100644 src/data/data.py
create mode 100644 src/data/data_preproc.py
create mode 100644 src/models/find_components/Find_Component.py
create mode 100644 src/models/find_components/__init__.py
create mode 100644 src/models/fit_methods/Fit_Methods.py
create mode 100644 src/models/fit_methods/__init__.py
create mode 100644 src/models/fit_methods/model.py
create mode 100644 src/utils/__init__.py
create mode 100644 src/utils/utils.py
diff --git a/README.md b/README.md
index 89637b7..f76d3e1 100644
--- a/README.md
+++ b/README.md
@@ -2,3 +2,14 @@
Aim of the project to transform the data from PES to SPEC.
+# Usage
+
+1. inv_train.py -> Train a model on the specific RUN.
+Thish will save the pca model and fit model in experiments/YOUR_DIR/checkpoints and the data in
+experiments/YOUR_DIR/data.
+
+2. inv_eval.py -> Use to evaluate the trained model
+
+3. inv_inference -> Use trained model to do inference on new data point
+
+4. data_drift_check.py -> Check data drift between two datasets
\ No newline at end of file
diff --git a/data_drift_check.py b/data_drift_check.py
new file mode 100644
index 0000000..6bef495
--- /dev/null
+++ b/data_drift_check.py
@@ -0,0 +1,34 @@
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+
+from src.data.data import ReadPesSpec, PesChannelSelector
+from src.data.data_preproc import SpecPreprocessing
+
+from src.models.find_components.Find_Component import FindPCAcomps
+from src.models.fit_methods.Fit_Methods import FitBFGS, FitBFGSPCA
+from src.models.fit_methods.model import Model
+
+from src.utils.utils import load_train_test_h5, load_rec_data_h5, load_rec_data_h5_bfgs_pca, load_checkpoint
+from src.utils.utils import create_experiment_dirs
+
+from sklearn.model_selection import train_test_split
+import numpy as np
+
+
+# Load trained model in the path exp_dir
+exp_dir = "test3_pulseen_short_test_eps_r0015"
+Y_train_model, Y_test_model, spec_train_model, spec_test_model, spec_raw_pe, X_train_model, X_test_model, pes_train_model, pes_test_model, att_dict, xgm_pulseen_train, xgm_pulseen_test = load_train_test_h5(exp_dir)
+
+att_dict["pes_pca_preprocessing"] = False # set pca non trainable
+att_dict["spec_pca_preprocessing"] = False # set pca non trainable
+
+model_instance = Model(model_type="bfgs_pca_eps", data_info=att_dict)
+
+inference_model, pes_pca_model, spec_pca_model = model_instance.load_model() # Move to incference sctipt TODO!
+
+print("Finish")
\ No newline at end of file
diff --git a/inv_eval.py b/inv_eval.py
new file mode 100644
index 0000000..8985125
--- /dev/null
+++ b/inv_eval.py
@@ -0,0 +1,1627 @@
+
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+
+from src.utils.utils import load_train_test_h5, load_rec_data_h5, load_rec_data_h5_bfgs_pca, load_checkpoint, load_rec_data_h5_bfgs_pca_eps
+import numpy as np
+import h5py
+
+exp_dir = "test3_pulseen_short_test_eps_r0015" # Evaluate the trained model
+
+# Load Data
+Y_train, Y_test, spec_train, spec_test, spec_raw_pe, X_train, X_test, pes_train, pes_test, att_dict, xgm_pulseen_trains, xgm_pulseen_tests = load_train_test_h5(exp_dir)
+
+xgm_pulseen_test = []
+for xgm_i in range(len(xgm_pulseen_tests)):
+ xgm_pulseen_test.append(xgm_pulseen_tests[xgm_i][0])
+
+n_pca_comps = att_dict["n_pca_comps_pes"]
+
+if att_dict["model_type"]=="bfgs":
+
+
+ # Load Rec
+ Y_rec, Y_rec_unc = load_rec_data_h5(exp_dir)
+ print(Y_test.shape, Y_rec.shape)
+
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+
+ for i_bfgs in range(15):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(np.arange(0,Y_rec.shape[1],1), Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(np.arange(0,Y_rec.shape[1],1), Y_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ np.arange(0,Y_rec.shape[1],1), Y_rec[i_bfgs,:] - Y_rec_unc, Y_rec[i_bfgs,:] + Y_rec_unc, color="pink", alpha=0.5, label="BFGS std"
+ )
+ ax.set_xlabel('XXX ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+if att_dict["model_type"]=="bfgs_pca":
+
+ # Load Rec
+ Y_rec, Y_rec_unc_bfgs, Y_rec_unc_specpca, Y_rec_unc = load_rec_data_h5_bfgs_pca(exp_dir)
+ print(Y_test.shape, Y_rec.shape)
+
+ Y_eps = 0*Y_rec_unc_bfgs # TODO! fix this remove eps version from here
+ eps_mean = np.mean(Y_eps, axis=1)
+ print("len eps mean", len(eps_mean))
+
+ # compute RMSE
+ def int_sum_1(A):
+ return np.sum(A, axis=1)
+ pes_sum = int_sum_1(pes_test)
+ spec_sum = int_sum_1(spec_test)
+
+ cor_mat = np.corrcoef(pes_test)
+ cor_mat_int = np.sum(cor_mat, axis=1)
+ #where_1 = np.where(cor_mat == 1)
+ #cor_mat[where_1] = 0
+
+
+
+ # quick fft eval
+ FC_START = 0
+ FC_END = 80
+ N_ROI = 40
+ # Get the Fourier Components
+ n_Y_comps = spec_test.shape[1]
+ Y_test_fft_raw = np.abs(np.fft.fft(spec_test))[:, :round(n_Y_comps/2)]
+ Y_rec_fft_raw = np.abs(np.fft.fft(Y_rec))[:, :round(n_Y_comps/2)]
+
+ # Cut the Fourier components and take only those who are relevant and not close to 0
+ Y_test_fft = Y_test_fft_raw[:, FC_START:FC_END]
+ Y_rec_fft = Y_rec_fft_raw[:, FC_START:FC_END]
+
+ # Split relevant fourier components into N regions of interests (roi)
+ split_size = round((FC_END - FC_START)/N_ROI)
+
+ def split_to_rois(a, splitedSize = split_size):
+ a_splited = [a[:,x:x+splitedSize] for x in range(0, a.shape[1], splitedSize)]
+ return a_splited
+
+ Y_test_fft_rois = split_to_rois(Y_test_fft)
+ Y_rec_fft_rois = split_to_rois(Y_rec_fft)
+
+ # Define the X axis, pixels
+ pixel_rois = split_to_rois(np.arange(FC_START, FC_END, 1).reshape(1,-1))
+
+ # Compute the absolute difference of gt and rec fourier components per region of interest per train ID
+ delta_Y_fft_rois = [np.abs(a-b) for a, b in zip(Y_test_fft_rois, Y_rec_fft_rois)]
+ # Sum along all Fourier components in roi per train id
+ delta_Y_fft_rois_sum = [np.sum(a, axis=1) for a in delta_Y_fft_rois]
+ print("len(delta_Y_fft_rois_sum)", len(delta_Y_fft_rois))
+ # Sum along all train IDs per roi
+ delta_Y_fft_rois_sum_mean = [x.mean() for x in delta_Y_fft_rois_sum]
+
+
+ # find corr coef
+ def corr_coef(a, b):
+ cor_coef = np.corrcoef(a, b)[0][1]
+ #print("The corr coef is :", np.round(cor_coef, 2))
+ return np.round(cor_coef, 2)
+ cc = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ #corr_coef(sc_res[i_bfgs,:], Y_test[i_bfgs,:])
+ cc.append(corr_coef(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]) )
+
+
+
+ # Function to calculate Chi-distance
+ def chi2_distance(A, B):
+ # compute the chi-squared distance using above formula
+ chi = 0.5 * np.sum([((a - b) ** 2) / (a + b)
+ for (a, b) in zip(A, B)])
+
+ #print("The Chi-square distance is :", chi)
+ return chi
+ chi_2 = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ chi_2.append(chi2_distance(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]))
+
+
+ # compute RMSE
+ def rmse_value(A, B):
+ return np.sqrt(np.mean((A-B)**2))
+ rmse = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ rmse.append(rmse_value(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]))
+ rmse_quickfft = []
+ for i_bfgs in range(Y_rec_fft.shape[0]):
+ rmse_quickfft.append(rmse_value(Y_rec_fft[i_bfgs,:], Y_test_fft[i_bfgs,:]))
+
+
+
+
+
+
+ # Save Images
+
+ # Save cor mat test set
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ im = ax.imshow(cor_mat)
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('train IDs.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor mat: PCA(PES)", y=1.0, pad=-20, fontsize=22)
+
+ cax = fig.add_axes([0.27, 0.95, 0.5, 0.05])
+ fig.colorbar(im, cax=cax, orientation='horizontal')
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_mat_pca_pes.png", bbox_inches='tight')
+ plt.close(fig)
+
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ ax.plot(cor_mat[42,:], label=f"42")
+ ax.plot(cor_mat[43,:], label=f"43")
+ ax.plot(cor_mat[171,:], label=f"171")
+ #ax.plot(cor_mat_int, label=f"int")
+ ax.legend(fontsize=20)
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_mat_pca_pes_line.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(xgm_pulseen_test)), xgm_pulseen_test, label=f"xgm pulse energy: PES")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"xgm test pulse energy: PES", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/xgm_pulse_energy.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs PES sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, xgm_pulseen_test, label=f"eps_mean xgm pulse energy")
+
+ ax.set_xlabel('eps_mean', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, rmse_quickfft, label=f"eps_mean QuickFFT")
+
+ ax.set_xlabel('eps_mean', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"eps_mean QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/eps_mean_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(xgm_pulseen_test, rmse_quickfft, label=f"xgm_pulseen_test QuickFFT")
+
+ ax.set_xlabel('xgm_pulseen_test', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pulse energy QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/xgm_pulseen_test_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save CC vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(cc, rmse_quickfft, label=f"cc vs QuickFFT")
+
+ ax.set_xlabel('cc', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc vs QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs Spec sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(spec_sum, xgm_pulseen_test, label=f"xgm pulse energy vs spec int")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_spec_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs corr matrix sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(cor_mat_int, xgm_pulseen_test, label=f"xgm pulse energy vs cor_mat_int")
+
+ ax.set_xlabel('cor_mat_int', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_cor_mat_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # pes_sum vs Spec sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_sum, pes_sum, label=f" spec int vs pes_sum")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pes_sum", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_vs_pes_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes sum test data
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(pes_sum)), pes_sum, label=f"integrated int: PES")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"integrated int: PES", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_int_sum.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes sum test data
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(spec_sum)), spec_sum, label=f"integrated int: SPEC")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"integrated int: SPEC", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_sum.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes_sum vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ #ax.plot(np.arange(len(cc)), pes_sum/max(np.array(pes_sum)), label=f"pes_sum")
+
+ ax.plot(pes_sum/max(np.array(pes_sum)), cc/max(cc), label=f"correlation coef.")
+
+ ax.set_xlabel('pes_sum', fontsize=22)
+ ax.set_ylabel('cor. coef.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes_sum vs cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_int_sum_vs_cor_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+ # Save spec_sum vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ #ax.plot(np.arange(len(cc)), spec_sum/max(np.array(spec_sum)), label=f"spec_sum")
+
+ ax.plot(spec_sum/max(np.array(spec_sum)), cc/max(cc), label=f"spec_sum correlation coef")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('cor. coef.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_sum_vs_corr_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save quick fft eval
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(N_ROI), delta_Y_fft_rois_sum_mean, label=f"FFT eval {i_bfgs}")
+
+ ax.set_xlabel('FFT bin comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f" qucik fft eval", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/quick_fft_eval.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(cc)), cc, label=f"correlation coef")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/corr_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ ax.scatter( cc/max(cc), eps_mean/spec_sum, label=f"cc vs eps_mean")
+
+ ax.set_xlabel('cor coef', fontsize=22)
+ ax.set_ylabel('eps mean', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc vs eps_mean", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_coef_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ ax.scatter( cc/max(cc), xgm_pulseen_test/max(np.array(xgm_pulseen_test)), label=f"xgm_pulseen_test")
+
+ ax.set_xlabel('cor coef', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor. coef. vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_coef_vs_pulse_energy.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmax(cc)
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc best rec", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_best_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmin(cc)
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc worse rec", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_worse_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec pes
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmax(cc)
+
+ ax.plot(pes_test[i_bfgs,:], "r", label=f"PES test {i_bfgs}")
+
+ ax.set_xlabel(' x comps ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes raw (cc best rec)", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_raw_best_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation worst rec pes
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmin(cc)
+
+ ax.plot(pes_test[i_bfgs,:], "r", label=f"PES test {i_bfgs}")
+
+ ax.set_xlabel(' x comps ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes raw (cc worst rec)", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_raw_worst_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save chi_2
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(chi_2)), chi_2, label=f"chi_2")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"chi_2", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/chi_2.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save rmse
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(rmse)), rmse, label=f"rmse")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"rmse", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/rmse.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save rmse vs eps_mean
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, rmse, label=f"rmse_vs_eps_mean")
+
+ ax.set_xlabel('eps mean', fontsize=22)
+ ax.set_ylabel('rmse', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"rmse_vs_eps_mean", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/rmse_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # reconstruction with total unc
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc[0,:], Y_rec[i_bfgs,:] + Y_rec_unc[0,:], color="pink", alpha=0.5, label="u_total"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir} eps={eps_mean[i_bfgs]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_eps[0,:], Y_rec[i_bfgs,:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir} eps={eps_mean[i_bfgs]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ print(Y_rec_unc_bfgs[0,:])
+ # Reconstruction with 2 uncs
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_2unc.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Single ID reconstruction with 2 unc
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_2unc_id1.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Single ID reconstruction with TOTAL unc
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc[0,:], Y_rec[i_bfgs,:] + Y_rec_unc[0,:], color="r", alpha=0.5, label="u_total"
+ )
+
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_unc_id1.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_spec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single Y
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(Y_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('PCA comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_Y.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single PES
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(pes_train[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('Channel comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_pes.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # # Example single X
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(X_train[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('PCA comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_X.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ args_cor_mat_int = np.argsort(eps_mean)
+ args_cor_mat_int = args_cor_mat_int[230:]
+ # ORDERED reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+ i_bfgs = i_bfgs
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:], "r", label=f"SPEC rec {i_bfgs, args_cor_mat_int[i_bfgs]}")
+ ax.plot(spec_raw_pe, spec_test[args_cor_mat_int[i_bfgs],:], label=f"SPEC gt {i_bfgs, args_cor_mat_int[i_bfgs]}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:] - Y_rec_unc_specpca, Y_rec[args_cor_mat_int[i_bfgs],:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:] - Y_eps[0,:], Y_rec[args_cor_mat_int[i_bfgs],:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"eps={eps_mean[args_cor_mat_int[i_bfgs]]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps_SORTED.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ args_eps = np.argsort(eps_mean)
+ # ORDERED reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[args_eps[i_bfgs],:], "r", label=f"SPEC rec {i_bfgs, args_eps[i_bfgs]}")
+ ax.plot(spec_raw_pe, spec_test[args_eps[i_bfgs],:], label=f"SPEC gt {i_bfgs, args_eps[i_bfgs]}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_eps[i_bfgs],:] - Y_rec_unc_specpca, Y_rec[args_eps[i_bfgs],:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_eps[i_bfgs],:] - Y_eps[0,:], Y_rec[args_eps[i_bfgs],:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"eps={eps_mean[args_eps[i_bfgs]]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps_SORTED_descending.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+
+
+
+ print("finished eval")
+
+
+if att_dict["model_type"]=="bfgs_pca_eps":
+
+ # Load Rec
+ Y_rec, Y_rec_unc_bfgs, Y_rec_unc_specpca, Y_rec_unc, Y_eps = load_rec_data_h5_bfgs_pca_eps(exp_dir)
+ print(Y_test.shape, Y_rec.shape)
+
+ eps_mean = np.mean(Y_eps, axis=1)
+ eps_mean = np.round(eps_mean, 2) # round eps mean
+ print("len eps mean", len(eps_mean))
+
+ # compute RMSE
+ def int_sum_1(A):
+ return np.sum(A, axis=1)
+ pes_sum = int_sum_1(pes_test)
+ spec_sum = int_sum_1(spec_test)
+
+ cor_mat = np.corrcoef(pes_test)
+ cor_mat_int = np.sum(cor_mat, axis=1)
+ #where_1 = np.where(cor_mat == 1)
+ #cor_mat[where_1] = 0
+
+
+
+ # quick fft eval
+ FC_START = 0
+ FC_END = 80
+ N_ROI = 40
+ # Get the Fourier Components
+ n_Y_comps = spec_test.shape[1]
+ Y_test_fft_raw = np.abs(np.fft.fft(spec_test))[:, :round(n_Y_comps/2)]
+ Y_rec_fft_raw = np.abs(np.fft.fft(Y_rec))[:, :round(n_Y_comps/2)]
+
+ # Cut the Fourier components and take only those who are relevant and not close to 0
+ Y_test_fft = Y_test_fft_raw[:, FC_START:FC_END]
+ Y_rec_fft = Y_rec_fft_raw[:, FC_START:FC_END]
+
+ # Split relevant fourier components into N regions of interests (roi)
+ split_size = round((FC_END - FC_START)/N_ROI)
+
+ def split_to_rois(a, splitedSize = split_size):
+ a_splited = [a[:,x:x+splitedSize] for x in range(0, a.shape[1], splitedSize)]
+ return a_splited
+
+ Y_test_fft_rois = split_to_rois(Y_test_fft)
+ Y_rec_fft_rois = split_to_rois(Y_rec_fft)
+
+ # Define the X axis, pixels
+ pixel_rois = split_to_rois(np.arange(FC_START, FC_END, 1).reshape(1,-1))
+
+ # Compute the absolute difference of gt and rec fourier components per region of interest per train ID
+ delta_Y_fft_rois = [np.abs(a-b) for a, b in zip(Y_test_fft_rois, Y_rec_fft_rois)]
+ # Sum along all Fourier components in roi per train id
+ delta_Y_fft_rois_sum = [np.sum(a, axis=1) for a in delta_Y_fft_rois]
+ print("len(delta_Y_fft_rois_sum)", len(delta_Y_fft_rois))
+ # Sum along all train IDs per roi
+ delta_Y_fft_rois_sum_mean = [x.mean() for x in delta_Y_fft_rois_sum]
+
+
+ # find corr coef
+ def corr_coef(a, b):
+ cor_coef = np.corrcoef(a, b)[0][1]
+ #print("The corr coef is :", np.round(cor_coef, 2))
+ return np.round(cor_coef, 2)
+ cc = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ #corr_coef(sc_res[i_bfgs,:], Y_test[i_bfgs,:])
+ cc.append(corr_coef(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]) )
+
+
+
+ # Function to calculate Chi-distance
+ def chi2_distance(A, B):
+ # compute the chi-squared distance using above formula
+ chi = 0.5 * np.sum([((a - b) ** 2) / (a + b)
+ for (a, b) in zip(A, B)])
+
+ #print("The Chi-square distance is :", chi)
+ return chi
+ chi_2 = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ chi_2.append(chi2_distance(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]))
+
+
+ # compute RMSE
+ def rmse_value(A, B):
+ return np.sqrt(np.mean((A-B)**2))
+ rmse = []
+ for i_bfgs in range(Y_rec.shape[0]):
+ rmse.append(rmse_value(Y_rec[i_bfgs,:], spec_test[i_bfgs,:]))
+ rmse_quickfft = []
+ for i_bfgs in range(Y_rec_fft.shape[0]):
+ rmse_quickfft.append(rmse_value(Y_rec_fft[i_bfgs,:], Y_test_fft[i_bfgs,:]))
+
+
+
+
+
+
+ # Save Images
+
+ # Save cor mat test set
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ im = ax.imshow(cor_mat)
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('train IDs.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor mat: PCA(PES)", y=1.0, pad=-20, fontsize=22)
+
+ cax = fig.add_axes([0.27, 0.95, 0.5, 0.05])
+ fig.colorbar(im, cax=cax, orientation='horizontal')
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_mat_pca_pes.png", bbox_inches='tight')
+ plt.close(fig)
+
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ ax.plot(cor_mat[42,:], label=f"42")
+ ax.plot(cor_mat[43,:], label=f"43")
+ ax.plot(cor_mat[171,:], label=f"171")
+ #ax.plot(cor_mat_int, label=f"int")
+ ax.legend(fontsize=20)
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_mat_pca_pes_line.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(xgm_pulseen_test)), xgm_pulseen_test, label=f"xgm pulse energy: PES")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"xgm test pulse energy: PES", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/xgm_pulse_energy.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs PES sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, xgm_pulseen_test, label=f"eps_mean xgm pulse energy")
+
+ ax.set_xlabel('eps_mean', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, rmse_quickfft, label=f"eps_mean QuickFFT")
+
+ ax.set_xlabel('eps_mean', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"eps_mean QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/eps_mean_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(xgm_pulseen_test, rmse_quickfft, label=f"xgm_pulseen_test QuickFFT")
+
+ ax.set_xlabel('xgm_pulseen_test', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pulse energy QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/xgm_pulseen_test_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save CC vs QuickFFT
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(cc, rmse_quickfft, label=f"cc vs QuickFFT")
+
+ ax.set_xlabel('cc', fontsize=22)
+ ax.set_ylabel('QuickFFT', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc vs QuickFFT", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_vs_quick_fft.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs Spec sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(spec_sum, xgm_pulseen_test, label=f"xgm pulse energy vs spec int")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_spec_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save xgm_pulse energy vs corr matrix sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(cor_mat_int, xgm_pulseen_test, label=f"xgm pulse energy vs cor_mat_int")
+
+ ax.set_xlabel('cor_mat_int', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pulse_energy_vs_cor_mat_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # pes_sum vs Spec sum
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_sum, pes_sum, label=f" spec int vs pes_sum")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs pes_sum", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_vs_pes_int.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes sum test data
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(pes_sum)), pes_sum, label=f"integrated int: PES")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"integrated int: PES", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_int_sum.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes sum test data
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(spec_sum)), spec_sum, label=f"integrated int: SPEC")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"integrated int: SPEC", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_sum.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save pes_sum vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ #ax.plot(np.arange(len(cc)), pes_sum/max(np.array(pes_sum)), label=f"pes_sum")
+
+ ax.plot(pes_sum/max(np.array(pes_sum)), cc/max(cc), label=f"correlation coef.")
+
+ ax.set_xlabel('pes_sum', fontsize=22)
+ ax.set_ylabel('cor. coef.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes_sum vs cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_int_sum_vs_cor_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+ # Save spec_sum vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ #ax.plot(np.arange(len(cc)), spec_sum/max(np.array(spec_sum)), label=f"spec_sum")
+
+ ax.plot(spec_sum/max(np.array(spec_sum)), cc/max(cc), label=f"spec_sum correlation coef")
+
+ ax.set_xlabel('spec_sum', fontsize=22)
+ ax.set_ylabel('cor. coef.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"spec_sum vs cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/spec_int_sum_vs_corr_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save quick fft eval
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(N_ROI), delta_Y_fft_rois_sum_mean, label=f"FFT eval {i_bfgs}")
+
+ ax.set_xlabel('FFT bin comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f" qucik fft eval", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/quick_fft_eval.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(cc)), cc, label=f"correlation coef")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor. coef.", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/corr_coef.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ ax.scatter( cc/max(cc), eps_mean/spec_sum, label=f"cc vs eps_mean")
+
+ ax.set_xlabel('cor coef', fontsize=22)
+ ax.set_ylabel('eps mean', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc vs eps_mean", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_coef_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation vs pulse energy
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ #ax.plot(np.arange(len(cc)), cc/max(cc), label=f"cross correlation")
+ ax.scatter( cc/max(cc), xgm_pulseen_test/max(np.array(xgm_pulseen_test)), label=f"xgm_pulseen_test")
+
+ ax.set_xlabel('cor coef', fontsize=22)
+ ax.set_ylabel('pulse energy', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cor. coef. vs pulse energy", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cor_coef_vs_pulse_energy.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmax(cc)
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc best rec", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_best_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmin(cc)
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"cc worse rec", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/cc_worse_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation best rec pes
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmax(cc)
+
+ ax.plot(pes_test[i_bfgs,:], "r", label=f"PES test {i_bfgs}")
+
+ ax.set_xlabel(' x comps ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes raw (cc best rec)", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_raw_best_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save cross correlation worst rec pes
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ i_bfgs = np.argmin(cc)
+
+ ax.plot(pes_test[i_bfgs,:], "r", label=f"PES test {i_bfgs}")
+
+ ax.set_xlabel(' x comps ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"pes raw (cc worst rec)", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/pes_raw_worst_rec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save chi_2
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(chi_2)), chi_2, label=f"chi_2")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"chi_2", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/chi_2.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save rmse
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(np.arange(len(rmse)), rmse, label=f"rmse")
+
+ ax.set_xlabel('train IDs.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"rmse", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/rmse.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Save rmse vs eps_mean
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.scatter(eps_mean, rmse, label=f"rmse_vs_eps_mean")
+
+ ax.set_xlabel('eps mean', fontsize=22)
+ ax.set_ylabel('rmse', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ ax.set_title(f"rmse_vs_eps_mean", y=1.0, pad=-20, fontsize=22)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/rmse_vs_eps_mean.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # reconstruction with total unc
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc[0,:], Y_rec[i_bfgs,:] + Y_rec_unc[0,:], color="pink", alpha=0.5, label="u_total"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir} eps={eps_mean[i_bfgs]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_eps[0,:], Y_rec[i_bfgs,:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir} eps={eps_mean[i_bfgs]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ print(Y_rec_unc_bfgs[0,:])
+ # Reconstruction with 2 uncs
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"{exp_dir}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_2unc.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Single ID reconstruction with 2 unc
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_bfgs[0,:], Y_rec[i_bfgs,:] + Y_rec_unc_bfgs[0,:], color="r", alpha=0.5, label="u_bfgs"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc_specpca, Y_rec[i_bfgs,:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_2unc_id1.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Single ID reconstruction with TOTAL unc
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, Y_rec[i_bfgs,:], "r", label=f"SPEC rec {i_bfgs}")
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[i_bfgs,:] - Y_rec_unc[0,:], Y_rec[i_bfgs,:] + Y_rec_unc[0,:], color="r", alpha=0.5, label="u_total"
+ )
+
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_unc_id1.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single spec
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(spec_raw_pe, spec_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_spec.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single Y
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(Y_test[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('PCA comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_Y.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # Example single PES
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(pes_train[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('Channel comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_pes.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ # # Example single X
+ fig, ax = plt.subplots(1, 1, figsize=(20, 10))
+ for i_bfgs in range(1):
+
+ ax.plot(X_train[i_bfgs,:], label=f"SPEC gt {i_bfgs}")
+
+ ax.set_xlabel('PCA comps.', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.tick_params(axis="y", labelsize=20)
+ ax.legend(fontsize=20)
+ #ax.set_title(f"{exp_dir}", y=1.0, pad=-20)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/single_X.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ args_cor_mat_int = np.argsort(eps_mean)
+ args_cor_mat_int = args_cor_mat_int[230:]
+ # ORDERED reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+ i_bfgs = i_bfgs
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:], "r", label=f"SPEC rec {i_bfgs, args_cor_mat_int[i_bfgs]}")
+ ax.plot(spec_raw_pe, spec_test[args_cor_mat_int[i_bfgs],:], label=f"SPEC gt {i_bfgs, args_cor_mat_int[i_bfgs]}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:] - Y_rec_unc_specpca, Y_rec[args_cor_mat_int[i_bfgs],:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_cor_mat_int[i_bfgs],:] - Y_eps[0,:], Y_rec[args_cor_mat_int[i_bfgs],:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"eps={eps_mean[args_cor_mat_int[i_bfgs]]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps_SORTED.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+ args_eps = np.argsort(eps_mean)
+ # ORDERED reconstruction with total unc + eps
+ fig, axes_2 = plt.subplots(10, 2, figsize=(15, 35))
+ axes_2 = axes_2.flatten()
+ for i_bfgs in range(20):
+
+ ax = axes_2[i_bfgs]
+ ax.plot(spec_raw_pe, Y_rec[args_eps[i_bfgs],:], "r", label=f"SPEC rec {i_bfgs, args_eps[i_bfgs]}")
+ ax.plot(spec_raw_pe, spec_test[args_eps[i_bfgs],:], label=f"SPEC gt {i_bfgs, args_eps[i_bfgs]}")
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_eps[i_bfgs],:] - Y_rec_unc_specpca, Y_rec[args_eps[i_bfgs],:] + Y_rec_unc_specpca, color="g", alpha=0.5, label="u_specpca"
+ )
+ ax.fill_between(
+ spec_raw_pe, Y_rec[args_eps[i_bfgs],:] - Y_eps[0,:], Y_rec[args_eps[i_bfgs],:] + Y_eps[0,:], color="blue", alpha=0.5, label="eps"
+ )
+ ax.set_xlabel('Energy (eV) ', fontsize=22)
+ ax.set_ylabel('int a.u.', fontsize=22)
+ ax.tick_params(axis="x", labelsize=20)
+ ax.legend(fontsize=10)
+ ax.set_title(f"eps={eps_mean[args_eps[i_bfgs]]}", y=1.0, pad=-10)
+
+ fig.savefig(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/summaries/reconstruction_eps_SORTED_descending.png", bbox_inches='tight')
+ plt.close(fig)
+
+
+
+
+
+ print("finished eval")
+
+
+
+
+
+#import joblib
+#spec_pca_model = joblib.load(f"/home/adavtyan/my_repos/invasive/experiments/{exp_dir}/checkpoints/spec_pca_model.joblib")
+
+
diff --git a/inv_inference.py b/inv_inference.py
new file mode 100644
index 0000000..6ad195c
--- /dev/null
+++ b/inv_inference.py
@@ -0,0 +1,69 @@
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+
+from src.data.data import ReadPesSpec, PesChannelSelector
+from src.data.data_preproc import SpecPreprocessing
+
+from src.models.find_components.Find_Component import FindPCAcomps
+from src.models.fit_methods.Fit_Methods import FitBFGS, FitBFGSPCA
+from src.models.fit_methods.model import Model
+
+from src.utils.utils import load_train_test_h5, load_rec_data_h5, load_rec_data_h5_bfgs_pca, load_checkpoint
+from src.utils.utils import create_experiment_dirs
+
+from sklearn.model_selection import train_test_split
+import numpy as np
+
+
+# Load trained model in the path exp_dir
+exp_dir = "test3_pulseen_short_test_eps_r0015"
+Y_train_model, Y_test_model, spec_train_model, spec_test_model, spec_raw_pe, X_train_model, X_test_model, pes_train_model, pes_test_model, att_dict, xgm_pulseen_train, xgm_pulseen_test = load_train_test_h5(exp_dir)
+
+att_dict["pes_pca_preprocessing"] = False # set pca non trainable
+att_dict["spec_pca_preprocessing"] = False # set pca non trainable
+
+
+# Change the exp_dir to new dir where inference data is located
+#att_dict["exp_dir"] = "test3_inference"
+att_dict["run_number"] = "r0014"
+att_dict["spec_ofset"] = -2
+
+
+RPS = ReadPesSpec(att_dict) # read data
+data_dict = RPS.get_data()
+
+
+model_instance = Model(model_type="bfgs_pca_eps", data_info=att_dict)
+pes_train, pes_test, spec_train, spec_test = model_instance.preprocess(data_dict, att_dict) # Model Preprocess
+
+
+inference_model, pes_pca_model, spec_pca_model = model_instance.load_model() # Move to incference sctipt TODO!
+
+
+# Update the Train and Test data for inference model
+model_instance.X_train, model_instance.X_test = pes_pca_model.transform(pes_train), pes_pca_model.transform(pes_test)
+model_instance.Y_train, model_instance.Y_test = spec_pca_model.transform(spec_train), spec_pca_model.transform(spec_test)
+
+
+inf_dir = exp_dir + "_inference_TestDATA" + att_dict["run_number"]
+model_instance.data_info["exp_dir"] = inf_dir
+
+
+create_experiment_dirs(inf_dir)
+model_instance.save_preproc()
+
+
+result = model_instance.predict(input_value=model_instance.X_test,
+ target_value=model_instance.Y_test,
+ spec_target=spec_test_model)
+
+
+model_instance.save_prediction()
+
+
+print("Finished Inference")
\ No newline at end of file
diff --git a/inv_train.py b/inv_train.py
new file mode 100644
index 0000000..17c693d
--- /dev/null
+++ b/inv_train.py
@@ -0,0 +1,103 @@
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+import matplotlib
+matplotlib.use('Agg')
+import matplotlib.pyplot as plt
+
+from src.data.data import ReadPesSpec, PesChannelSelector
+from src.data.data_preproc import SpecPreprocessing
+
+from src.models.find_components.Find_Component import FindPCAcomps
+
+from src.models.fit_methods.model import Model
+
+from src.utils.utils import create_experiment_dirs, save_train_test_h5, save_rec_data_h5, save_checkpoint, save_checkpoint_fft
+
+from sklearn.model_selection import train_test_split
+import numpy as np
+
+run_directory = "/gpfs/exfel/exp/SA3/202121/p002935/raw"
+run_number = "r0015"
+spec_ofset = -2
+n_pca_comps_pes = 400
+n_pca_comps_spec = 20
+pes_pca_preprocessing = True
+spec_pca_preprocessing = True
+spec_gc = True
+spec_fft = False
+use_data_subset = True
+data_subset_start, data_subset_end = 1000, 6000 # Min Max
+channel_names = "all_chs" # "default", "all_chs", or list with channel names ex ["channel_1_D", "channel_2_B"]
+test_size = 0.05
+model_type = "bfgs_pca_eps" # "bfgs", "bfgs_pca", "bfgs_pca_eps"
+
+exp_dir = "test3_pulseen_short_test_eps_" + run_number
+general_comment = exp_dir
+
+data_info = {"run_directory": run_directory,
+ "run_number": run_number,
+ "spec_ofset": spec_ofset,
+ "exp_dir": exp_dir,
+ "n_pca_comps_pes": n_pca_comps_pes,
+ "n_pca_comps_spec": n_pca_comps_spec,
+ "pes_pca_preprocessing": pes_pca_preprocessing,
+ "spec_pca_preprocessing": spec_pca_preprocessing,
+ "spec_gc": spec_gc,
+ "spec_fft": spec_fft,
+ "use_data_subset": use_data_subset,
+ "data_subset_start": data_subset_start,
+ "data_subset_end": data_subset_end,
+ "channel_names": channel_names,
+ "test_size": test_size,
+ "model_type": model_type,
+ "general_comment": general_comment
+
+ }
+
+
+experiment_dir, summary_dir, checkpoint_dir, output_dir, test_dir = create_experiment_dirs(exp_dir)
+
+import datetime
+now = datetime.datetime.now()
+print ("Start Read Pes Spec Data : ")
+print (now.strftime("%Y-%m-%d %H:%M:%S"))
+
+# Read Pes Spec Data
+RPS = ReadPesSpec(data_info)
+data_dict = RPS.get_data()
+
+now = datetime.datetime.now()
+print ("End Read Pes Spec Data : ")
+print (now.strftime("%Y-%m-%d %H:%M:%S"))
+
+# Model
+model_instance = Model(model_type=model_type, data_info=data_info)
+X_train, X_test, Y_train, Y_test = model_instance.preprocess(data_dict, data_info) # Model Preprocess
+
+now = datetime.datetime.now()
+print ("End Model Preprocessor : ")
+print (now.strftime("%Y-%m-%d %H:%M:%S"))
+
+model_instance.save_preproc()
+print("################")
+print(X_train.shape)
+print("################")
+model_instance.fit_eval(X_train, Y_train, X_test, Y_test) # Model FIT
+
+now = datetime.datetime.now()
+print ("End Model Train : ")
+print (now.strftime("%Y-%m-%d %H:%M:%S"))
+
+model_instance.save_latest_ckp()
+
+result = model_instance.predict(input_value=X_test)
+
+model_instance.save_prediction()
+
+
+
+
+print("FINISHED ALL")
+
diff --git a/src/data/__init__.py b/src/data/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/data/data.py b/src/data/data.py
new file mode 100644
index 0000000..cd9517e
--- /dev/null
+++ b/src/data/data.py
@@ -0,0 +1,326 @@
+"""
+ Data loaders for SPEC and PES.
+"""
+
+from extra_data import open_run, RunDirectory
+import numpy as np
+
+from typing import Any, Union, List, Dict
+
+import os
+import h5py
+import numpy as np
+
+import itertools
+
+import scipy
+import scipy.interpolate
+from random import choices
+
+from skimage.feature import canny
+import torch
+import torchvision.transforms as transforms
+from collections import namedtuple
+import torch.nn.functional as F
+
+import cv2
+
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+
+from sklearn.datasets import fetch_olivetti_faces
+from numpy.random import RandomState
+
+
+class ReadPesSpec:
+ """
+ Read PES and SPEC data, select by unique IDs
+ """
+
+ def __init__(self, data_info):
+
+ self.data_info = data_info
+
+ self.rundir = self.data_info["run_directory"]
+ self.runid = self.data_info["run_number"]
+ self.spec_ofset = self.data_info["spec_ofset"]
+
+ self.run = RunDirectory(f"{self.rundir}/{self.runid}")
+
+ self.spec_inds = None
+ self.pes_inds = None
+ self.xgm_inds = None
+
+ self.spec_tid_all = None
+ self.spec_tid_all_offset = None
+ self.pes_tid_all = None
+ self.xgm_tid_all = None
+
+ self.spec_tids = None
+ self.pes_tids = None
+ self.xgm_tids = None
+
+ self.retvol_raw = None
+ self.retvol_raw_timestamp = None
+ self.min_ret_vol = None
+
+
+ self.spec_raw_pe = None
+ self.spec_raw_int = None
+ self.pes_raw_dict = None
+ self.pes_data = None
+ self.xgm_pressureFiltered = None
+ self.xgm_pulseen = None
+ self.retvol_raw = None
+ self.raw_timestamp = None
+ self.min_ret_vol = None
+
+
+ # Output
+ self.data = {"spec_raw_pe":None,
+ "spec_raw_int":None,
+ "pes_raw_dict":None,
+ "pes_data":None,
+ "xgm_pressureFiltered":None,
+ "xgm_pulseen":None,
+ "retvol_raw":None,
+ "retvol_raw_timestamp":None,
+ "min_ret_vol":None}
+
+
+
+ def get_unq_inds(self, a, b, c):
+ uniq_vals = list(set(a).intersection(b).intersection(c))
+ return [a.index(x) for x in uniq_vals], [b.index(x) for x in uniq_vals], [c.index(x) for x in uniq_vals]
+
+ # create channel names
+ def gen_channel_names(self):
+ channels = []
+ for i in range(4):
+ channels.append(f"channel_{i+1}_A")
+ channels.append(f"channel_{i+1}_B")
+ channels.append(f"channel_{i+1}_C")
+ channels.append(f"channel_{i+1}_D")
+
+ chs_short_name = []
+ for i in range(4):
+ chs_short_name.append(f"ch_{i+1}_A")
+ chs_short_name.append(f"ch_{i+1}_B")
+ chs_short_name.append(f"ch_{i+1}_C")
+ chs_short_name.append(f"ch_{i+1}_D")
+
+ return channels, chs_short_name
+
+
+ def read_train_ids(self, debug=False):
+
+ # Read raw SPEC Data train IDs
+ self.spec_tid_all = self.run['SA3_XTD10_SPECT/MDL/FEL_BEAM_SPECTROMETER_SQS1:output', f"data.trainId"].ndarray()
+ self.spec_tid_all_offset = self.spec_tid_all + self.spec_ofset ##### Offset correction #####
+ #print(f" spec_tid_all_offset = {len(spec_tid_all_offset)}")
+
+ # Read raw PES Data train IDs
+ self.pes_tid_all = self.run['SA3_XTD10_PES/ADC/1:network', f"digitizers.trainId"].ndarray()
+ #print(f" pes_tid_all = {len(pes_tid_all)}")
+
+
+ # Read raw XGM Data train IDs
+ self.xgm_tid_all = self.run['SA3_XTD10_XGM/XGM/DOOCS:output', f"data.trainId"].ndarray()
+ #print(f" xgm_tid_all = {len(xgm_tid_all)}")
+
+ print(f" Introduced SPEC_OFFSET = {self.spec_ofset} \n ")
+
+ if debug:
+ print(f"\n Len of train IDs: \n \
+ spec_tid_all_offset {len(self.spec_tid_all_offset)} \n \
+ pes_tid_all {len(self.pes_tid_all)} \n \
+ xgm_tid_all {len(self.xgm_tid_all)} \n \
+ ")
+
+
+ print(f"\n First 5 train IDs from all sources: \n \
+ spec_tid_all {self.spec_tid_all[:5]} \n \
+ spec_tid_all_offset {self.spec_tid_all_offset[:5]} \n \
+ pes_tid_all {self.pes_tid_all[:5]} \n \
+ xgm_tid_all {self.xgm_tid_all[:5]} \n \
+ ")
+
+
+ def merge_train_ids(self, debug=True):
+ # Merge Train IDs
+ self.spec_inds, self.pes_inds, self.xgm_inds = self.get_unq_inds(list(self.spec_tid_all_offset), list(self.pes_tid_all), list(self.xgm_tid_all))
+
+
+ print(" Len of indices for common elements in spec, pes, xgm :", len(self.spec_inds), len(self.pes_inds), len(self.xgm_inds))
+
+ # Find common train ids
+ self.spec_tids = self.spec_tid_all_offset[self.spec_inds]
+ self.pes_tids = self.pes_tid_all[self.pes_inds]
+ self.xgm_tids = self.xgm_tid_all[self.xgm_inds]
+
+ if debug:
+ print(f"\n First 5 indices from all sources: \n \
+ spec_inds {self.spec_inds[:5]} \n \
+ pes_inds {self.pes_inds[:5]} \n \
+ xgm_inds {self.xgm_inds[:5]} \n \
+ ")
+
+ print(f"\n First 5 train IDs from all sources: \n \
+ spec_tids {self.spec_tids[:5]} \n \
+ pes_tids {self.pes_tids[:5]} \n \
+ xgm_tids {self.xgm_tids[:5]} \n \
+ ")
+ return self.spec_inds, self.pes_inds, self.xgm_inds
+
+
+ def retVol(self):
+ # Read PES Data RetVol
+ self.retvol_raw = self.run["SA3_XTD10_PES/MDL/DAQ_MPOD", "u212.value"].ndarray()[list(self.pes_inds)]
+ self.min_ret_vol = np.abs(self.retvol_raw.min())
+ self.retvol_raw_timestamp = self.run["SA3_XTD10_PES/MDL/DAQ_MPOD", "u212.timestamp"].ndarray()[list(self.pes_inds)]
+
+
+ def read_data(self, debug=True):
+
+ # Read raw SPEC Data
+ self.spec_raw_pe= self.run['SA3_XTD10_SPECT/MDL/FEL_BEAM_SPECTROMETER_SQS1:output', f"data.photonEnergy"].ndarray().astype(float)[list(self.spec_inds),:]
+ #spec_raw_all = run_spec['SA3_XTD10_SPECT/MDL/FEL_BEAM_SPECTROMETER_SQS1:output', f"data.intensityDistribution"].ndarray()
+ self.spec_raw_int = self.run['SA3_XTD10_SPECT/MDL/FEL_BEAM_SPECTROMETER_SQS1:output', f"data.intensityDistribution"].ndarray().astype(float)[list(self.spec_inds),:]
+
+ # Read PES Data
+ channels, _ = self.gen_channel_names() # create channel names
+ self.pes_raw_dict = {}
+ pes_raw_dict_all = {}
+ for ch in channels:
+ #pes_raw = run['SA3_XTD10_PES/ADC/1:network', f"digitizers.{ch}.raw.samples"].ndarray()
+ #pes_raw_dict_all[ch] = run_pes['SA3_XTD10_PES/ADC/1:network', f"digitizers.{ch}.raw.samples"].ndarray()
+ self.pes_raw_dict[ch] = self.run['SA3_XTD10_PES/ADC/1:network', f"digitizers.{ch}.raw.samples"].ndarray().astype(float)[list(self.pes_inds),:]
+
+ # Read XGM gas pressure
+ self.xgm_pressureFiltered= self.run['SA3_XTD10_XGM/XGM/DOOCS', f"pressure.pressureFiltered.value"].ndarray().astype(float)[list(self.xgm_inds)]
+
+ # Read PES Data RetVol
+ self.retVol()
+
+ # Read XGM Pusle Energy
+ self.xgm_pulseen = self.run['SA3_XTD10_XGM/XGM/DOOCS:output', f"data.intensitySa3TD"].ndarray().astype(float)[list(self.xgm_inds)]
+
+ if debug:
+ #print(f"\n pes data channels :: \n {self.pes_raw_dict.keys()}")
+ print(f"spec_raw_pe.shape{self.spec_raw_pe.shape}, \n spec_raw_int.shape{self.spec_raw_int.shape}")
+ print(f"\n pes channel shape channel_1_A:: {self.pes_raw_dict['channel_1_A'].shape}")
+ print("\n shape of xgm_pressureFiltered", self.xgm_pressureFiltered.shape)
+ print(f"\n min_ret_vol = {self.min_ret_vol}")
+
+
+ assert self.spec_raw_pe.shape, self.spec_raw_int.shape
+ assert self.pes_raw_dict['channel_1_A'].shape[0], self.spec_raw_int.shape[0]
+
+
+
+ def get_data(self):
+
+ # Read Train Ids
+ self.read_train_ids(self.run)
+
+ # Merge Train Ids
+ self.merge_train_ids()
+
+ # Read Data
+ self.read_data()
+
+ self.data["spec_raw_pe"] = self.spec_raw_pe
+ self.data["spec_raw_int"] = self.spec_raw_int
+ self.data["pes_raw_dict"] = self.pes_raw_dict
+ self.data["xgm_pressureFiltered"] = self.xgm_pressureFiltered
+ self.data["xgm_pulseen"] = self.xgm_pulseen
+ self.data["retvol_raw"] = self.retvol_raw
+ self.data["retvol_raw_timestamp"] = self.retvol_raw_timestamp
+ self.data["min_ret_vol"] = self.min_ret_vol
+
+ print("Data loaded in dataframe")
+
+ PCS = PesChannelSelector(pes_dict=self.data, ch_names=self.data_info["channel_names"]) # Select Pes Channels
+ self.pes_data = PCS.sel_pes_chs()
+ self.data["pes_data"] = self.pes_data
+
+ print("pes_data.shape = ", self.pes_data.shape) # (14497, 1200)
+ print("spec_data.shape = ", self.spec_raw_int.shape) # (14497, 1900)
+
+
+ return self.data
+
+
+class PesChannelSelector:
+ """_Use this class to take all or part of PES channels for data preprocessing_
+ """
+ def __init__(self, pes_dict, ch_names="default"):
+
+ self.pes_dict = pes_dict
+ self.default_channels = ["channel_1_D", "channel_2_B", "channel_3_A", "channel_3_B", "channel_4_C", "channel_4_D"]
+
+
+ if ch_names is "default":
+ self.channels = self.default_channels
+ elif ch_names == "all_chs":
+ self.channels, _ = self.gen_channel_names()
+ else:
+ self.channels = ch_names
+
+ # Output
+ self.pes_data = np.array([])
+
+
+
+ # create channel names
+ def gen_channel_names(self):
+ """
+ Generate channel names for quick reference
+
+ Returns:
+ [channels], [chs_short_name]: raw channel names, channel short names
+ """
+ channels = []
+ for i in range(4):
+ channels.append(f"channel_{i+1}_A")
+ channels.append(f"channel_{i+1}_B")
+ channels.append(f"channel_{i+1}_C")
+ channels.append(f"channel_{i+1}_D")
+
+ chs_short_name = []
+ for i in range(4):
+ chs_short_name.append(f"ch_{i+1}_A")
+ chs_short_name.append(f"ch_{i+1}_B")
+ chs_short_name.append(f"ch_{i+1}_C")
+ chs_short_name.append(f"ch_{i+1}_D")
+
+ return channels, chs_short_name
+
+
+ def get_array_from_dict(self, ch_name):
+ """ Select ROI where TOF is present
+
+ Args:
+ ch_name (str): channel name
+
+ Returns:
+ Croped channel : crop channel around the TOF measurement
+ """
+ TOF_start = 31445 + 0
+ TOF_end = TOF_start + 200
+ return self.pes_dict["pes_raw_dict"][ch_name][:,TOF_start:TOF_end]
+
+
+ def sel_pes_chs(self):
+ """ Select PES channel
+
+ Returns:
+ pes_data: PES data concat together all channels
+ """
+
+ self.pes_data = np.concatenate([self.get_array_from_dict(ch_name) for ch_name in self.channels], axis=1)
+
+ return self.pes_data
+
diff --git a/src/data/data_preproc.py b/src/data/data_preproc.py
new file mode 100644
index 0000000..fba4bb0
--- /dev/null
+++ b/src/data/data_preproc.py
@@ -0,0 +1,99 @@
+"""
+ Data Preprocessing
+"""
+
+import scipy
+import scipy.signal
+import numpy as np
+from sklearn import preprocessing
+
+import sys
+sys.path.append("./")
+sys.path.append("..")
+
+from src.models.find_components.Find_Component import FindPCAcomps
+
+
+# GC Spec data
+class SpecPreprocessing():
+ def __init__(self, spec_data, spec_raw_pe, gauss_sigma=0.2):
+
+ self.gauss_sigma = gauss_sigma
+ self.spec_data = spec_data
+ self.spec_raw_pe = spec_raw_pe
+
+ # Output
+ self.gaussian = None
+ self.spec_data_gc = None
+ self.spec_data_gc_fft = None
+
+ def gaussian_convolve(self):
+ gx = self.spec_raw_pe[:,:]
+ mu = self.spec_raw_pe[0,gx.shape[1]//2]
+
+ self.gaussian = np.exp(-((gx-mu)/self.gauss_sigma)**2/2)/np.sqrt(2*np.pi*self.gauss_sigma**2)
+ print(self.spec_data.shape, self.spec_raw_pe.shape, self.gaussian.shape)
+ #result = np.convolve(spec_data[0,:], gaussian, mode="full")
+
+ self.spec_data_gc = scipy.signal.fftconvolve(self.spec_data, self.gaussian, mode="same",axes=1)/80
+ print(self.spec_data_gc.shape)
+
+ return self.spec_data_gc, self.gaussian
+
+ def make_fft(self):
+
+ self.spec_data_gc_fft = np.fft.fft(self.spec_data_gc)
+
+ return self.spec_data_gc_fft
+
+
+
+class PesPreprocessing():
+ def __init__(self, data_train, data_test, n_pca=50):
+
+ self.data_train = data_train
+ self.data_test = data_test
+ self.n_pca = n_pca
+
+ # Output
+ self.pca_model = None
+ self.scaler = None
+ self.new_train = None
+ self.new_test = None
+
+
+
+ def find_pca(self):
+ pes_PCA = FindPCAcomps(
+ data_train=self.data_train,
+ data_test=self.data_test,
+ n_pca_comps=self.n_pca
+ )
+
+ self.pca_model, self.new_train, self.new_test = pes_PCA.get_pca()
+ return self.pca_model, self.new_train, self.new_test
+
+ def stand_scaler(self):
+
+ self.scaler = preprocessing.StandardScaler().fit(self.data_train)
+
+ self.new_train = self.scaler.transform(self.data_train)
+ self.new_test = self.scaler.transform( self.data_test)
+
+ return self.scaler, self.new_train, self.new_test
+
+
+ def get(self, data_info):
+
+ if data_info["pes_pca_preprocessing"]:
+ self.pca_model, self.new_train, self.new_test = self.find_pca()
+ return self.pca_model, self.new_train, self.new_test
+
+ if data_info["standard_scaler"]:
+ self.new_train, self.new_test = self.stand_scaler()
+ return self.scaler, self.new_train, self.new_test
+
+
+
+
+
diff --git a/src/models/find_components/Find_Component.py b/src/models/find_components/Find_Component.py
new file mode 100644
index 0000000..54c93b0
--- /dev/null
+++ b/src/models/find_components/Find_Component.py
@@ -0,0 +1,194 @@
+import sys
+sys.path.append("..")
+
+
+from sklearn.datasets import load_digits
+from sklearn.datasets import fetch_olivetti_faces
+from sklearn.decomposition import FastICA
+from sklearn import decomposition
+from sklearn.cluster import MiniBatchKMeans
+from sklearn import decomposition
+import matplotlib.pyplot as plt
+import skimage
+
+import numpy as np
+from numpy.random import RandomState
+
+import logging
+from time import time
+
+from sklearn import linear_model
+import torch
+
+from scipy import signal
+from scipy.signal import find_peaks
+from sklearn.decomposition import PCA
+
+class FindComponents(object):
+ """This class find the components in data. Those components could be independent or principal. It also returns the transformer.
+
+ Args:
+ object (object): Object of components and transformer
+ """
+
+
+ def __init__(self):
+ """_summary_
+
+ Args:
+ train_data (np.array): _description_
+ val_data (np.array): _description_
+ image_shape (tuple, optional): _description_. Defaults to (64, 64).
+ is_image (bool, optional): _description_. Defaults to True.
+ """
+
+ super(FindComponents, self).__init__()
+
+ self.transformer = None
+ self.ica_comps = None
+ self.ica_comps_pix = None
+
+ def clear_input(self):
+ self.train_data = None
+ self.val_data = None
+ self.image_shape = None
+ self.is_image = None
+
+
+ def get_transformer_ica(self, train_data, val_data, image_shape=(64, 64), is_image=True, N_COMP_ICA=5, ICA_Method='logcosh'):
+ """Takes number of data points and calculates the N Independent components based on sklearn.decomposition.FastICA.
+ Returns the transformer and corresponding ICA components in 2 verions (N ICA Components, and N ICA components per pixel.
+
+ Args:
+ train_data (np.array): _description_
+ val_data (np.array): _description_
+ image_shape (tuple, optional): _description_. Defaults to (64, 64).
+ is_image (bool, optional): _description_. Defaults to True.
+ N_COMP_ICA (int, optional): _description_. Defaults to 5.
+
+ Returns:
+ [transformer (object)]: transformer object for FastICA
+ [ica_comps (list)]: FastICA N componenrs ordered for all pixels in single data
+ [ica_comps_pix (list)]: FastICA N components ordered per pixel for all components in single data
+ """
+
+ self.train_data = train_data
+ self.val_data = val_data
+ self.image_shape = image_shape
+ self.is_image = is_image
+
+ self.train_data_points = self.train_data.shape[0]
+ if len(val_data.shape)==1:
+ self.val_data_points = 1
+ elif len(val_data.shape)!=1:
+ self.val_data_points = self.val_data.shape[0]
+
+ assert(N_COMP_ICA<=self.train_data_points)
+
+ transformer = FastICA(n_components=N_COMP_ICA, max_iter=5000, random_state=0, whiten=True, fun=ICA_Method)
+ print("shape train data", self.train_data.shape)
+ print("ICA_Method", ICA_Method)
+ X_transformed = transformer.fit_transform(self.train_data)
+ print("X_transformed.shape=", X_transformed.shape)
+
+ n_img, l_img = transformer.mixing_.T.shape
+ print("transformer.mixing_.T.shape", n_img, l_img)
+ ica_comps = [transformer.mixing_[:, i] for i in range(N_COMP_ICA)]
+ #print("ica_comps", ica_comps)
+
+
+ # Get pixel wise N components by
+ ica_comps_pix = [transformer.mixing_[i, :] for i in range(l_img)]
+ #print("ica_comps_pix", ica_comps_pix)
+
+ self.transformer = transformer
+ self.ica_comps = ica_comps
+ self.ica_comps_pix = ica_comps_pix
+
+ return self.transformer, self.ica_comps, self.ica_comps_pix, X_transformed
+
+
+class FindNumberOfComponents(object):
+
+ def __init__(self):
+ self.comps = None
+ self.PSDs = None
+ self.PSDs_means = None
+ self.user_n_components = None
+ self.X_centered, self.y_centered = None, None
+
+ self.n_good_comps = None
+ self.good_comps = None
+
+
+ def find_components(self):
+ # Find components
+ number_of_components = self.user_n_components
+ fit_method = "BayesianRidge"
+
+ FC = FindComponents()
+ transformer, comps, comps_pix, X_transformed = FC.get_transformer_ica(train_data=self.X_centered, val_data=self.y_centered, N_COMP_ICA=self.user_n_components) # finds components
+ self.comps = comps
+
+ def find_PSDs(self):
+ #self.find_components()
+ self.PSDs = [signal.welch(ac)[1] for ac in self.comps]
+ return self.PSDs
+
+ def find_peaks_PSD(self):
+ self.PSDs_means = np.mean(self.PSDs, axis=1)
+ peaks, _ = find_peaks(self.PSDs_means, threshold=self.PSDs_means.min()*3)
+ self.n_good_comps = len(peaks)
+ self.good_comps = peaks
+ print("DONE !!!!!!!!!!!!!!!!!!")
+ return self.n_good_comps, self.good_comps
+
+ def n_good_components(self,
+ X_centered,
+ y_centered,
+ number_of_components=20):
+
+ self.user_n_components = number_of_components
+ self.X_centered, self.y_centered = X_centered, y_centered
+
+ self.find_components()
+ self.find_PSDs()
+ self.find_peaks_PSD()
+
+ #plt.plot(self.PSDs_means)
+ #plt.plot(self.good_comps, self.PSDs_means[self.good_comps], "*")
+ #plt.plot(np.zeros_like(self.PSDs_means), "--", color="gray")
+ #plt.show()
+ #print(f"n_good_components = {self.n_good_comps}")
+ #print(f"Good components = {self.good_comps}")
+
+
+ return self.n_good_comps, self.good_comps
+
+
+class FindPCAcomps(object):
+ """_summary_
+
+ Args:
+ object (_type_): _description_
+ """
+
+ def __init__(self, data_train, data_test, n_pca_comps=50):
+ self.pca_model = None
+ self.data_train = data_train
+ self.data_test = data_test
+ self.n_pca_comps = n_pca_comps
+
+ def clear_output(self):
+ self.pca_model = None
+ self.pca_train = None
+ self.pca_test = None
+
+
+ def get_pca(self):
+ self.clear_output()
+ self.pca_model = PCA(n_components=self.n_pca_comps, whiten=True)
+ self.pca_train = self.pca_model.fit_transform(self.data_train)
+ self.pca_test = self.pca_model.transform(self.data_test)
+
+ return self.pca_model, self.pca_train, self.pca_test
\ No newline at end of file
diff --git a/src/models/find_components/__init__.py b/src/models/find_components/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/models/fit_methods/Fit_Methods.py b/src/models/fit_methods/Fit_Methods.py
new file mode 100644
index 0000000..e7baf72
--- /dev/null
+++ b/src/models/fit_methods/Fit_Methods.py
@@ -0,0 +1,302 @@
+from sklearn import linear_model
+import torch
+import scipy
+import h5py
+import numpy as np
+from autograd import numpy as anp # Thinly-wrapped version of Numpy
+from autograd import grad
+
+
+
+
+class FitBFGSPCA_EPS(object):
+
+ def __init__(self):
+
+ self.X_train = None
+ self.Y_train = None
+
+ self.X_test = None
+ self.Y_test = None
+
+ self.Y_train_norm = None
+ self.Y_test_norm = None
+
+ self.spec_test = None
+
+ self.Nx = None
+ self.Ny = None
+ self.A0 = None
+ self.Aeps = None
+
+ self.b0 = None
+ self.u0 = None
+ self.x0 = None
+
+ self.sc_op = None
+ self.A_inf = None
+ self.b_inf = None
+ self.u_inf = None
+
+ self.model = {"A_inf":self.A_inf,
+ "b_inf":self.b_inf,
+ "u_inf":self.u_inf
+
+ }
+
+ self.pca_model_spec = None
+
+ self.res = None
+ self.res_unc = None
+ self.res_pca_unc = None
+ self.res_unc_specpca = None
+
+ self.result = {"res":self.res,
+ "res_unc":self.res_unc,
+ "res_pca_unc": self.res_pca_unc,
+ "res_unc_specpca": self.res_unc_specpca
+ }
+
+ self.input_data = None
+ self.loss_train = None
+ self.loss_test = None
+
+
+ def loss_mae(self, x):
+ # diag( (in @ x - out) @ (in @ x - out)^T )
+ A = x[:self.Nx*self.Ny].reshape((self.Nx, self.Ny))
+ b = x[self.Nx*self.Ny:(self.Nx*self.Ny+self.Ny)].reshape((1, self.Ny))
+ log_unc = anp.clip(x[(self.Nx*self.Ny+self.Ny):].reshape((1, self.Ny)), -3, 8)
+ #log_unc = x[(Nx*Ny+Ny):].reshape((1, Ny))
+ iunc = anp.exp(-log_unc)
+ #print("iunc2", iunc2)
+ #print("log_unc", log_unc)
+
+ return anp.mean((anp.fabs(self.X_train @ A + b - self.Y_train)*iunc + log_unc).sum(axis=1), axis=0 ) # Put RELU on (XX@x) and introduce new matrix W
+
+ def loss_regul(self, x):
+ # diag( (in @ x - out) @ (in @ x - out)^T )
+ A = x[:self.Nx*self.Ny].reshape((self.Nx, self.Ny))
+ b = x[self.Nx*self.Ny:(self.Nx*self.Ny+self.Ny)].reshape((1, self.Ny))
+ #log_unc = anp.clip(x[(Nx*Ny+Ny):].reshape((1, Ny)), -3, 8)
+ log_unc = x[(self.Nx*self.Ny+self.Ny):].reshape((1, self.Ny))
+ iunc2 = anp.exp(-2*log_unc)
+ unc2 = anp.exp(2*log_unc)
+ #print("iunc2", iunc2)
+ #print("log_unc", log_unc)
+ width = 100.0
+ w = 0.5/width**2
+ #prior = w*((A**2).reshape((1, -1)).mean(axis=1))
+
+
+ # New from Danilo-> p(unc|a, b) = ba/gamma(a) * unc(a-1) * np.exp(-b * unc)
+ a = 3
+ b = 6
+ # prior -> - log p(unc|a, b) = - (a-1)*anp.log(unc) + b*unc
+ prior = - (a-1) * log_unc + b * anp.exp(log_unc)
+
+
+ return anp.mean( (0.5*((self.X_train @ A + b - self.Y_train_norm)**2)*iunc2 + log_unc).sum(axis=1) + prior.sum(axis=1), axis=0 ) # Put RELU on (XX@x) and introduce new matrix W
+
+ def fill_init(self):
+
+
+ self.Nx = self.X_train.shape[1]
+ self.Ny = self.Y_train.shape[1]
+ self.A0 = np.eye(self.Nx, self.Ny).reshape(self.Nx*self.Ny)
+ self.Aeps = np.zeros(self.Nx)
+
+ self.b0 = np.zeros(self.Ny)
+ self.u0 = np.zeros(self.Ny)
+ self.x0 = np.concatenate((self.A0,self.b0,self.u0,self.Aeps))
+
+
+
+
+ def fit_eval(self, X_train, Y_train, X_test, Y_test):
+
+ self.X_train = X_train
+ self.Y_train = Y_train
+
+ self.X_test = X_test
+ self.Y_test = Y_test
+
+ self.fill_init()
+
+
+ def loss(x, X, Y):
+ # diag( (in @ x - out) @ (in @ x - out)^T )
+ A = x[:self.Nx*self.Ny].reshape((self.Nx, self.Ny))
+ b = x[self.Nx*self.Ny:(self.Nx*self.Ny+self.Ny)].reshape((1, self.Ny))
+
+ b_eps = x[(self.Nx*self.Ny+self.Ny):(self.Nx*self.Ny+self.Ny+self.Ny)].reshape((1, self.Ny))
+
+ A_eps = x[(self.Nx*self.Ny+self.Ny+self.Ny):].reshape((self.Nx, 1))
+ log_unc = X @ A_eps + b_eps
+
+ #log_unc = anp.log(anp.exp(log_unc) + anp.exp(log_eps))
+ iunc2 = anp.exp(-2*log_unc)
+
+ #print("iunc2", iunc2)
+ #print("log_unc", log_unc)
+
+ return anp.mean( (0.5*((X@ A + b - Y)**2)*iunc2 + log_unc).sum(axis=1), axis=0 ) # Put RELU on (XX@x) and introduce new matrix W
+
+ self.loss_train = []
+ self.loss_test = []
+
+ def loss_hist(x):
+ l_train = loss(x, X_train, self.Y_train)
+ l_test = loss(x, X_test, self.Y_test)
+
+ self.loss_train.append(l_train)
+ self.loss_test.append(l_test)
+ return l_train
+
+ def loss_hist_2(x):
+ l_train = loss(x, self.X_train, self.Y_train)
+ return l_train
+
+ grad_loss = grad(loss_hist_2)
+ #self.grad_loss_regul = grad(loss_regul)
+ self.sc_op = scipy.optimize.fmin_l_bfgs_b(loss_hist, self.x0, grad_loss, disp=True, factr=1e7, maxiter=100, iprint = 0)
+
+ # Inference
+ #print(sc_res.shape)
+ self.A_inf = self.sc_op[0][:self.Nx*self.Ny].reshape(self.Nx, self.Ny)
+ self.b_inf = self.sc_op[0][self.Nx*self.Ny:(self.Nx*self.Ny+self.Ny)].reshape(1, self.Ny)
+ self.u_inf = self.sc_op[0][(self.Nx*self.Ny+self.Ny):(self.Nx*self.Ny+self.Ny+self.Ny)].reshape(1, self.Ny) # removed np.exp
+ self.A_eps = self.sc_op[0][(self.Nx*self.Ny+self.Ny+self.Ny):].reshape(self.Nx, 1)
+
+ self.model = {"A_inf":self.A_inf,
+ "b_inf":self.b_inf,
+ "u_inf":self.u_inf,
+ "A_eps":self.A_eps,
+ "loss_train": self.loss_train,
+ "loss_test": self.loss_test,
+ }
+
+ return self.model
+
+ def save_preproc(self,
+ pes_train, pes_test,
+ spec_train, spec_test,
+ X_train, X_test,
+ Y_train, Y_test,
+ spec_raw_pe,
+ data_info,
+ xgm_pulseen_train,
+ xgm_pulseen_test
+ ):
+
+ exp_dir = data_info["exp_dir"]
+
+ hf = h5py.File(f"experiments/{exp_dir}/output/data.h5", 'w')
+ hf.create_dataset('pes_train', data=pes_train)
+ hf.create_dataset('pes_test', data=pes_test)
+ hf.create_dataset('spec_train', data=spec_train)
+ hf.create_dataset('spec_test', data=spec_test)
+ hf.create_dataset('X_train', data=X_train)
+ hf.create_dataset('X_test', data=X_test)
+ hf.create_dataset('Y_train', data=Y_train)
+ hf.create_dataset('Y_test', data=Y_test)
+ hf.create_dataset('spec_raw_pe', data=spec_raw_pe[0,:])
+ hf.create_dataset('xgm_pulseen_train', data=xgm_pulseen_train)
+ hf.create_dataset('xgm_pulseen_test', data=xgm_pulseen_test)
+
+ for key, val in data_info.items():
+ hf.attrs[key] = val
+
+ def save_latest_chp(self, exp_dir):
+ filemodel = f"experiments/{exp_dir}/checkpoints/chp.h5"
+ hf = h5py.File(filemodel, 'w')
+
+ for k, v in self.model.items():
+ print("saving:", k)
+ hf.create_dataset(k, data=v)
+ hf.close()
+ print(f"model saved in -> {filemodel}")
+
+ def load_model(self, exp_dir):
+
+ # Load Checkpoint Data
+ print("loading model", exp_dir)
+ filename = f"experiments/{exp_dir}/checkpoints/chp.h5"
+ file = h5py.File(filename, 'r')
+ A_inf = file["A_inf"][()] # returns as a numpy array
+ b_inf = file["b_inf"][()] # returns as a numpy array
+ u_inf = file["u_inf"][()] # returns as a numpy array
+ A_eps = file["A_eps"][()] # returns as a numpy array
+
+
+ import joblib
+ pes_pca_model=joblib.load(f"experiments/{exp_dir}/checkpoints/pes_pca_model.joblib")
+ spec_pca_model=joblib.load(f"experiments/{exp_dir}/checkpoints/spec_pca_model.joblib")
+
+ self.model = {"A_inf":A_inf,
+ "b_inf":b_inf,
+ "u_inf":u_inf,
+ "A_eps":A_eps,
+ "pes_pca_model":pes_pca_model,
+ "spec_pca_model":spec_pca_model
+
+ }
+ return self.model, pes_pca_model, spec_pca_model
+
+ def predict(self, input_data, input_target, spec_target):
+
+ if self.pca_model_spec is not None:
+ self.model["spec_pca_model"] = self.pca_model_spec # TODO! check for training
+
+ if input_target is None:
+ self.model["Y_test"] = self.Y_test
+ else:
+ self.model["Y_test"] = input_target
+
+ if spec_target is None:
+ self.model["spec_target"] = self.spec_test
+ else:
+ self.model["spec_target"] = spec_target
+
+
+
+
+
+ self.result["res_pca"] = (input_data @ self.model["A_inf"] + self.model["b_inf"]) # inference reconstruction in PCA space
+ self.result["res_pca_unc"] = self.model["u_inf"][0,:] # unc in pca space
+ self.result["res_pca_eps"] = np.exp(input_data @ self.model["A_eps"] + self.result["res_pca_unc"]) # eps in pca space
+
+ self.result["res"] = self.model["spec_pca_model"].inverse_transform(self.result["res_pca"]) # transform PCA space to real space
+
+ self.result["res_unc"] = self.model["spec_pca_model"].inverse_transform(self.result["res_pca"] + self.model["u_inf"]) - self.model["spec_pca_model"].inverse_transform(self.result["res_pca"] )
+ self.result["res_unc"] = np.fabs(self.result["res_unc"])
+
+ self.result["res_eps"] = self.model["spec_pca_model"].inverse_transform(self.result["res_pca"] + self.result["res_pca_eps"]) - self.model["spec_pca_model"].inverse_transform(self.result["res_pca"] )
+ self.result["res_eps"] = np.fabs(self.result["res_eps"])
+
+
+ self.Yhat_pca = self.model["spec_pca_model"].inverse_transform(self.model["Y_test"])
+ self.result["res_unc_specpca"] = np.sqrt(((self.Yhat_pca - self.model["spec_target"])**2).mean(axis=0))
+
+ self.result["res_unc_total"] = np.sqrt(self.result["res_eps"]**2 + self.result["res_unc_specpca"]**2)
+ return self.result
+
+
+ def save_prediction(self, exp_dir):
+
+ fileprediction = f"experiments/{exp_dir}/output/pred_data.h5"
+ hf = h5py.File(fileprediction, 'w')
+ hf.create_dataset('res_pca', data=self.result["res_pca"])
+ hf.create_dataset('res_pca_unc', data=self.result["res_pca_unc"])
+ hf.create_dataset('res', data=self.result["res"])
+ hf.create_dataset('res_unc', data=self.result["res_unc"])
+ hf.create_dataset('res_unc_specpca', data=self.result["res_unc_specpca"])
+ hf.create_dataset('res_unc_total', data=self.result["res_unc_total"])
+ hf.create_dataset('res_pca_eps', data=self.result["res_pca_eps"])
+ hf.create_dataset('res_eps', data=self.result["res_eps"])
+
+ hf.close()
+
+ print(f"Prediction result saved in -> {fileprediction}")
+
+
diff --git a/src/models/fit_methods/__init__.py b/src/models/fit_methods/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/models/fit_methods/model.py b/src/models/fit_methods/model.py
new file mode 100644
index 0000000..db101ab
--- /dev/null
+++ b/src/models/fit_methods/model.py
@@ -0,0 +1,164 @@
+from sklearn.ensemble import RandomForestRegressor
+from sklearn.linear_model import LinearRegression
+from sklearn.model_selection import train_test_split
+import numpy as np
+
+import pandas as pd
+
+from src.models.fit_methods.Fit_Methods import FitBFGSPCA_EPS
+from src.data.data_preproc import SpecPreprocessing
+from src.models.find_components.Find_Component import FindPCAcomps
+
+from src.utils.utils import save_chp_bfgs
+
+
+class Model:
+ def __init__(self, model_type = None, data_info = None):
+
+
+ self.model_type = model_type
+ self.data_info = data_info
+
+ self.data_dict = None
+
+ self.pes_train = None
+ self.pes_test = None
+ self.spec_train = None
+ self.spec_test = None
+
+ self.xgm_pulseen = None
+ self.xgm_pulseen_train = None
+ self.xgm_pulseen_test = None
+
+ self.X_train = None
+ self.X_test = None
+ self.y_train = None
+ self.y_test = None
+
+ self.pes_pca_model = None
+ self.spec_pca_model = None
+
+
+ if self.model_type == 'bfgs':
+ self.user_defined_model = FitBFGS()
+ #self.user_defined_preproc = SpecPreprocessing()
+ elif self.model_type == 'bfgs_pca':
+ self.user_defined_model = FitBFGSPCA()
+ elif self.model_type == 'bfgs_pca_eps':
+ self.user_defined_model = FitBFGSPCA_EPS()
+
+
+
+ def split(self, test_size):
+ X = np.array(self.df[['Humidity', 'Pressure (millibars)']])
+ y = np.array(self.df['Temperature (C)'])
+ self.X_train, self.X_test, self.y_train, self.y_test = train_test_split(X, y, test_size = test_size, random_state = 42)
+
+ def preprocess(self, data_dict, data_info):
+ self.data_dict = data_dict
+ self.data_info = data_info
+
+ self.xgm_pulseen = data_dict["xgm_pulseen"]
+
+ self.spec_data = self.data_dict["spec_raw_int"].copy() # Spec Data Intensity
+ self.spec_data_pe = self.data_dict["spec_raw_pe"].copy() # Spec Data Photon energy
+ self.pes_data = self.data_dict["pes_data"].copy() # Pes Data Intensity
+
+
+ # Data Preprocessing
+ SP = SpecPreprocessing(self.spec_data, self.spec_data_pe)
+ self.spec_data_gc, _ = SP.gaussian_convolve()
+ self.spec_data = self.spec_data_gc #SP.make_fft()
+
+ # Split Data into TEST TRAIN
+ if self.data_info["use_data_subset"]:
+ self.pes_train, self.pes_test, self.spec_train, self.spec_test, self.xgm_pulseen_train, self.xgm_pulseen_test = train_test_split(self.pes_data[self.data_info["data_subset_start"]:self.data_info["data_subset_end"]],
+ self.spec_data[self.data_info["data_subset_start"]:self.data_info["data_subset_end"]],
+ self.xgm_pulseen[self.data_info["data_subset_start"]:self.data_info["data_subset_end"]],
+ test_size=self.data_info["test_size"],
+ random_state=42
+ )
+ else:
+ self.pes_train, self.pes_test, self.spec_train, self.spec_test, self.xgm_pulseen_train, self.xgm_pulseen_test = train_test_split(self.pes_data,
+ self.spec_data,
+ self.xgm_pulseen,
+ test_size=self.data_info["test_size"],
+ random_state=42
+ )
+
+
+ if self.data_info["pes_pca_preprocessing"]:
+ pes_PCA = FindPCAcomps(data_train=self.pes_train, data_test=self.pes_test, n_pca_comps=self.data_info["n_pca_comps_pes"])
+ self.pes_pca_model, self.X_train, self.X_test = pes_PCA.get_pca()
+ else:
+ self.X_train, self.X_test = self.pes_train, self.pes_test
+
+ if self.data_info["spec_pca_preprocessing"]:
+ spec_PCA = FindPCAcomps(data_train=self.spec_train, data_test=self.spec_test, n_pca_comps=self.data_info["n_pca_comps_spec"])
+ self.spec_pca_model, self.Y_train, self.Y_test = spec_PCA.get_pca()
+ self.user_defined_model.pca_model_spec = self.spec_pca_model
+ self.user_defined_model.spec_test = self.spec_test
+
+ else:
+ self.Y_train, self.Y_test = self.spec_train, self.spec_test
+
+ print("print Y shape train and test in preprocessing last step :::: ",self.Y_train.shape, self.Y_test.shape)
+ #save_train_test_h5(pes_train, pes_test, X_train, X_test, Y_train, Y_test, spec_data_pe, self.data_info, data_info["exp_dir"])
+ return self.X_train, self.X_test, self.Y_train, self.Y_test
+
+ def save_preproc(self):
+ if self.model_type == 'bfgs':
+
+ self.user_defined_model.save_preproc(self.pes_train, self.pes_test,
+ self.spec_train, self.spec_test,
+ self.X_train, self.X_test,
+ self.Y_train, self.Y_test,
+ self.spec_data_pe,
+ self.data_info,
+ )
+
+ import joblib
+ joblib.dump(self.pes_pca_model, f"experiments/{self.data_info['exp_dir']}/checkpoints/pes_pca_model.joblib")
+ joblib.dump(self.spec_pca_model, f"experiments/{self.data_info['exp_dir']}/checkpoints/spec_pca_model.joblib")
+ elif self.model_type == 'bfgs_pca' or self.model_type == 'bfgs_pca_eps':
+ self.user_defined_model.save_preproc(self.pes_train, self.pes_test,
+ self.spec_train, self.spec_test,
+ self.X_train, self.X_test,
+ self.Y_train, self.Y_test,
+ self.spec_data_pe,
+ self.data_info,
+ self.xgm_pulseen_train, self.xgm_pulseen_test
+ )
+
+ import joblib
+ joblib.dump(self.pes_pca_model, f"experiments/{self.data_info['exp_dir']}/checkpoints/pes_pca_model.joblib")
+ joblib.dump(self.spec_pca_model, f"experiments/{self.data_info['exp_dir']}/checkpoints/spec_pca_model.joblib")
+
+
+ def fit_eval(self, X_train, y_train, X_test, Y_test):
+ #A_inf, b_inf, u_inf, Y_norm = fig_bgfs.train_bgfs()
+ self.model = self.user_defined_model.fit_eval(X_train, y_train, X_test, Y_test) # Change the name of self.model to something else ?
+
+
+ def save_latest_ckp(self):
+ self.user_defined_model.save_latest_chp(self.data_info["exp_dir"])
+
+ def load_model(self):
+ self.model, self.pes_pca_model, self.spec_pca_model = self.user_defined_model.load_model(self.data_info["exp_dir"])
+ return self.model, self.pes_pca_model, self.spec_pca_model
+
+ def predict(self, input_value, target_value=None, spec_target=None):
+ result = self.user_defined_model.predict(input_value, target_value, spec_target)
+ return result
+
+ def save_prediction(self):
+ self.user_defined_model.save_prediction(self.data_info["exp_dir"])
+
+if __name__ == '__main__':
+ print("Creating Model Instance")
+ #model_instance = Model(model_type=None)
+ #model_instance.preprocess()
+ #model_instance.split(0.2)
+ #model_instance.fit()
+ #print(model_instance.predict([.9, 1000]))
+ #print("Accuracy: ", model_instance.model.score(model_instance.X_test, model_instance.y_test))
\ No newline at end of file
diff --git a/src/utils/__init__.py b/src/utils/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/utils/utils.py b/src/utils/utils.py
new file mode 100644
index 0000000..0db96e6
--- /dev/null
+++ b/src/utils/utils.py
@@ -0,0 +1,238 @@
+import numpy as np
+import scipy
+import matplotlib.pyplot as plt
+from skimage import filters as skfilters
+import plotly
+import plotly.graph_objs as go
+
+import os
+
+
+
+# plot lines
+def plot_gallery_line(title, data, n_col, n_row):
+ plt.figure(figsize=(2.0 * n_col, 2.26 * n_row))
+ plt.suptitle(title, size=16)
+ for i, comp in enumerate(data):
+ plt.subplot(n_row, n_col, i + 1)
+ vmax = max(comp.max(), -comp.min())
+ plt.plot(
+ comp
+ )
+
+ if i + 1 not in [k*n_col+1 for k in range(n_row)]:
+ plt.xticks(())
+ plt.yticks(())
+
+ plt.subplots_adjust(0.01, 0.05, 0.99, 0.93, 0.04, 0.0)
+
+
+
+def plot_unc_band(lp_hr, lp_rec, lp_rec_unc, fit_method):
+
+ x_pltly = np.arange(len(lp_hr))
+
+
+ fig = go.Figure([
+ go.Scatter(
+ name='HR (gt)',
+ x=x_pltly,
+ y=lp_hr,
+ mode='lines',
+ line=dict(color='rgb(0, 150, 255)'),
+ ),
+ go.Scatter(
+ name=f"{fit_method}",
+ x=x_pltly,
+ y=lp_rec,
+ mode='lines',
+ line=dict(color='rgb(255,99,71)'),
+ ),
+ go.Scatter(
+ name='Upper Bound',
+ x=x_pltly,
+ y=lp_rec + 2*lp_rec_unc,
+ mode='lines',
+ marker=dict(color="#DC143C"),
+ line=dict(width=0),
+ showlegend=False
+ ),
+ go.Scatter(
+ name='Lower Bound',
+ x=x_pltly,
+ y=lp_rec - 2*lp_rec_unc,
+ marker=dict(color="#DC143C"),
+ line=dict(width=0),
+ mode='lines',
+ fillcolor='rgba(255,99,71, 0.3)',
+ fill='tonexty',
+ showlegend=False
+ )
+ ])
+ fig.update_layout(
+ yaxis_title='log intensity (a.u.)',
+ xaxis_title='pixel/coord number (a.u.)',
+ title=f"{fit_method} reconstruction with 95% confidence on uncertainities",
+ hovermode="x"
+ )
+ fig.show()
+
+
+
+def create_experiment_dirs(exp_dir):
+ """
+ Create Directories of a regular tensorflow experiment directory
+ :param exp_dir:
+ :return summary_dir, checkpoint_dir:
+ """
+ experiment_dir = "experiments/" + exp_dir + "/"
+ summary_dir = experiment_dir + 'summaries/'
+ checkpoint_dir = experiment_dir + 'checkpoints/'
+ output_dir = experiment_dir + 'output/'
+ test_dir = experiment_dir + 'test/'
+ dirs = [summary_dir, checkpoint_dir, output_dir, test_dir]
+ try:
+ for dir_ in dirs:
+ if not os.path.exists(dir_):
+ os.makedirs(dir_)
+ print("Experiment directories created")
+ return experiment_dir, summary_dir, checkpoint_dir, output_dir, test_dir
+ except Exception as err:
+ print("Creating directories error: {0}".format(err))
+ exit(-1)
+
+
+def save_train_test_h5(pes_train, pes_test, X_train, X_test, Y_train, Y_test, spec_raw_pe, data_info, exp_dir):
+ import h5py
+
+ hf = h5py.File(f"experiments/{exp_dir}/output/data.h5", 'w')
+ hf.create_dataset('pes_train', data=pes_train)
+ hf.create_dataset('pes_test', data=pes_test)
+ hf.create_dataset('X_train', data=X_train)
+ hf.create_dataset('X_test', data=X_test)
+ hf.create_dataset('Y_train', data=Y_train)
+ hf.create_dataset('Y_test', data=Y_test)
+ hf.create_dataset('spec_raw_pe', data=spec_raw_pe[0,:])
+ for key, val in data_info.items():
+ hf.attrs[key] = val
+
+
+ hf.close()
+
+def save_rec_data_h5(sc_res, sc_res_unc, exp_dir):
+ import h5py
+ hf = h5py.File(f"experiments/{exp_dir}/output/rec_data.h5", 'w')
+ hf.create_dataset('sc_res', data=sc_res)
+ hf.create_dataset('sc_res_unc', data=sc_res_unc)
+ hf.close()
+
+
+def save_checkpoint(A_inf, b_inf, u_inf, Y_norm, exp_dir):
+
+ import h5py
+ hf = h5py.File(f"experiments/{exp_dir}/checkpoints/chp.h5", 'w')
+ hf.create_dataset('A_inf', data=A_inf)
+ hf.create_dataset('b_inf', data=b_inf)
+ hf.create_dataset('u_inf', data=u_inf)
+ hf.create_dataset('Y_norm', data=Y_norm)
+ hf.close()
+
+def save_chp_bfgs(model, exp_dir):
+ import h5py
+ hf = h5py.File(f"experiments/{exp_dir}/checkpoints/chp.h5", 'w')
+
+ for k, v in model.items():
+ hf.create_dataset(k, data=v)
+ hf.close()
+
+
+
+def save_checkpoint_fft(A_inf_1, b_inf_1,
+ A_inf_2, b_inf_2,
+ u_inf_1, u_inf_2,
+ Y_norm, exp_dir):
+
+ import h5py
+ hf = h5py.File(f"experiments/{exp_dir}/checkpoints/chp.h5", 'w')
+ hf.create_dataset('A_inf_1', data=A_inf_1)
+ hf.create_dataset('b_inf_1', data=b_inf_1)
+ hf.create_dataset('A_inf_2', data=A_inf_2)
+ hf.create_dataset('b_inf_2', data=b_inf_2)
+ hf.create_dataset('u_inf_1', data=u_inf_1)
+ hf.create_dataset('u_inf_2', data=u_inf_2)
+ hf.create_dataset('Y_norm', data=Y_norm)
+ hf.close()
+
+
+
+def load_train_test_h5(exp_dir):
+ import h5py
+ # Load Data
+ filename = f"experiments/{exp_dir}/output/data.h5"
+ file = h5py.File(filename, 'r')
+ pes_train = file["pes_train"][()] # returns as a numpy array
+ pes_test = file["pes_test"][()] # returns as a numpy array
+ spec_train = file["spec_train"][()] # returns as a numpy array
+ spec_test = file["spec_test"][()] # returns as a numpy array
+ X_train = file["X_train"][()] # returns as a numpy array
+ X_test = file["X_test"][()] # returns as a numpy array
+ Y_train = file["Y_train"][()] # returns as a numpy array
+ Y_test = file["Y_test"][()] # returns as a numpy array
+ spec_raw_pe = file["spec_raw_pe"][()] # returns as a numpy
+ xgm_pulseen_train = file["xgm_pulseen_train"][()]
+ xgm_pulseen_test = file["xgm_pulseen_test"][()]
+
+ att_names = file.attrs.keys()
+ att_dict = {}
+ for att_name in att_names:
+ att_dict[att_name] = file.attrs[att_name]
+
+ return Y_train, Y_test, spec_train, spec_test, spec_raw_pe, X_train, X_test, pes_train, pes_test, att_dict, xgm_pulseen_train, xgm_pulseen_test
+
+
+def load_rec_data_h5(exp_dir):
+ import h5py
+ filename = f"experiments/{exp_dir}/output/pred_data.h5"
+ file = h5py.File(filename, 'r')
+ sc_res = file["res"][()] # returns as a numpy array
+ sc_res_unc = file["res_unc"][()] # returns as a numpy array
+
+ return sc_res, sc_res_unc
+
+def load_rec_data_h5_bfgs_pca(exp_dir):
+ import h5py
+ filename = f"experiments/{exp_dir}/output/pred_data.h5"
+ file = h5py.File(filename, 'r')
+ sc_res = file["res"][()] # returns as a numpy array
+ sc_res_unc = file["res_unc"][()] # returns as a numpy array
+ sc_res_unc_specpca = file["res_unc_specpca"][()] # returns as a numpy array
+ sc_res_unc_total = file["res_unc_total"][()] # returns as a numpy array
+
+
+ return sc_res, sc_res_unc, sc_res_unc_specpca, sc_res_unc_total
+
+def load_rec_data_h5_bfgs_pca_eps(exp_dir):
+ import h5py
+ filename = f"experiments/{exp_dir}/output/pred_data.h5"
+ file = h5py.File(filename, 'r')
+ sc_res = file["res"][()] # returns as a numpy array
+ sc_res_unc = file["res_unc"][()] # returns as a numpy array
+ sc_res_unc_specpca = file["res_unc_specpca"][()] # returns as a numpy array
+ sc_res_unc_total = file["res_unc_total"][()] # returns as a numpy array
+ sc_res_eps = file["res_eps"][()] # returns as a numpy array
+
+ return sc_res, sc_res_unc, sc_res_unc_specpca, sc_res_unc_total, sc_res_eps
+
+def load_checkpoint(exp_dir):
+ import h5py
+ # Load Checkpoint Data
+ filename = f"experiments/{exp_dir}/checkpoints/chp.h5"
+ file = h5py.File(filename, 'r')
+ A_inf = file["A_inf"][()] # returns as a numpy array
+ b_inf = file["b_inf"][()] # returns as a numpy array
+ u_inf = file["u_inf"][()] # returns as a numpy array
+ Y_norm = file["Y_norm"][()] # returns as a numpy array
+
+ return A_inf, b_inf, u_inf, Y_norm
+
+
--
GitLab