From 6b5af7b136b29b7e4b2ee5c974432c6ceb4828c7 Mon Sep 17 00:00:00 2001
From: Danilo Ferreira de Lima <danilo.enoque.ferreira.de.lima@xfel.de>
Date: Wed, 21 Dec 2022 14:18:49 +0100
Subject: [PATCH] Record time taken for test. Remove unnecessary variables when
writing model to HDF5 file.
---
pes_to_spec/model.py | 52 +++++++---------------------
pes_to_spec/test/offline_analysis.py | 31 +++++++++++++----
2 files changed, 36 insertions(+), 47 deletions(-)
diff --git a/pes_to_spec/model.py b/pes_to_spec/model.py
index 98627fd..767c575 100644
--- a/pes_to_spec/model.py
+++ b/pes_to_spec/model.py
@@ -128,7 +128,7 @@ class Model(TransformerMixin, BaseEstimator):
channels:List[str]=[f"channel_{j}_{k}"
for j, k in product(range(1, 5), ["A", "B", "C", "D"])],
n_pca_lr: int=600,
- n_pca_hr: int=40,
+ n_pca_hr: int=20,
high_res_sigma: float=0.2,
tof_start: Optional[int]=None,
delta_tof: Optional[int]=300,
@@ -284,7 +284,7 @@ class Model(TransformerMixin, BaseEstimator):
Returns: Smoothened high resolution spectrum.
"""
- self.high_res_photon_energy = high_res_photon_energy
+ self.high_res_photon_energy = high_res_photon_energy[0, np.newaxis, :]
# if the prompt peak has not been given, guess it
if self.tof_start is None:
@@ -403,26 +403,14 @@ class FitModel(object):
Linear regression model with uncertainties.
"""
def __init__(self):
- # training dataset
- self.X_train: Optional[np.ndarray] = None
- self.Y_train: Optional[np.ndarray] = None
-
- # test dataset to evaluate uncertainty
- self.X_test: Optional[np.ndarray] = None
- self.Y_test: Optional[np.ndarray] = None
-
- # normalized target
- self.Y_train_norm = None
- self.Y_test_norm = None
-
# model parameter sizes
self.Nx: int = 0
self.Ny: int = 0
# fit result
- self.A_inf: np.ndarray = None
- self.b_inf: np.ndarray = None
- self.u_inf: np.ndarray = None
+ self.A_inf: Optional[np.ndarray] = None
+ self.b_inf: Optional[np.ndarray] = None
+ self.u_inf: Optional[np.ndarray] = None
# fit monitoring
self.loss_train: List[float] = list()
@@ -435,17 +423,9 @@ class FitModel(object):
Perform the fit and evaluate uncertainties with the test set.
"""
- # training dataset
- self.X_train: np.ndarray = X_train
- self.Y_train: np.ndarray = Y_train
-
- # test dataset to evaluate uncertainty
- self.X_test: np.ndarray = X_test
- self.Y_test: np.ndarray = Y_test
-
# model parameter sizes
- self.Nx: int = int(self.X_train.shape[1])
- self.Ny: int = int(self.Y_train.shape[1])
+ self.Nx: int = int(X_train.shape[1])
+ self.Ny: int = int(Y_train.shape[1])
# initial parameter values
A0: np.ndarray = np.eye(self.Nx, self.Ny).reshape(self.Nx*self.Ny)
@@ -491,8 +471,8 @@ class FitModel(object):
Returns: The loss value.
"""
- l_train = loss(x, self.X_train, self.Y_train)
- l_test = loss(x, self.X_test, self.Y_test)
+ l_train = loss(x, X_train, Y_train)
+ l_test = loss(x, X_test, Y_test)
self.loss_train += [l_train]
self.loss_test += [l_test]
@@ -507,7 +487,7 @@ class FitModel(object):
Returns: The loss value.
"""
- l_train = loss(x, self.X_train, self.Y_train)
+ l_train = loss(x, X_train, Y_train)
return l_train
grad_loss = grad(loss_train)
@@ -534,10 +514,6 @@ class FitModel(object):
Returns: Dictionary with all relevant variables.
"""
return dict(
- X_train=self.X_train,
- X_test=self.X_test,
- Y_train=self.Y_train,
- Y_test=self.Y_test,
A_inf=self.A_inf,
b_inf=self.b_inf,
u_inf=self.u_inf,
@@ -554,10 +530,6 @@ class FitModel(object):
in_dict: The input dictionary with relevant variables.
"""
- self.X_train = in_dict["X_train"]
- self.X_test = in_dict["X_test"]
- self.Y_train = in_dict["Y_train"]
- self.Y_test = in_dict["Y_test"]
self.A_inf = in_dict["A_inf"]
self.b_inf = in_dict["b_inf"]
self.u_inf = in_dict["u_inf"]
@@ -587,9 +559,9 @@ class FitModel(object):
result["Y_eps"] = np.exp(X @ self.A_eps + result["Y_unc"])
#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"] = 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"] = 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))
diff --git a/pes_to_spec/test/offline_analysis.py b/pes_to_spec/test/offline_analysis.py
index 9ce3272..2a165ce 100755
--- a/pes_to_spec/test/offline_analysis.py
+++ b/pes_to_spec/test/offline_analysis.py
@@ -17,6 +17,9 @@ from matplotlib.gridspec import GridSpec
from typing import Optional
+from time import time_ns
+import pandas as pd
+
def plot_pes(filename: str, pes_raw_int: np.ndarray):
"""
Plot low-resolution spectrum.
@@ -109,6 +112,9 @@ def main():
#retvol_raw = run["SA3_XTD10_PES/MDL/DAQ_MPOD", "u212.value"].select_trains(by_id[tids]).ndarray()
#retvol_raw_timestamp = run["SA3_XTD10_PES/MDL/DAQ_MPOD", "u212.timestamp"].select_trains(by_id[tids]).ndarray()
+ t = list()
+ t_names = list()
+
# these have been manually selected:
#useful_channels = ["channel_1_D",
# "channel_2_B",
@@ -116,34 +122,45 @@ def main():
# "channel_3_B",
# "channel_4_C",
# "channel_4_D"]
- model = Model(channels=channels,
- n_pca_lr=600,
- n_pca_hr=40,
- high_res_sigma=0.2,
- tof_start=None,
- delta_tof=300,
- validation_size=0.05)
+ model = Model()
train_idx = np.isin(tids, train_tids)
model.debug_peak_finding(pes_raw, "test_peak_finding.png")
print("Fitting")
+ start = time_ns()
model.fit({k: v[train_idx, :]
for k, v in pes_raw.items()},
spec_raw_int[train_idx, :],
spec_raw_pe[train_idx, :])
+ t += [time_ns() - start]
+ t_names += ["Fit"]
spec_smooth = model.preprocess_high_res(spec_raw_int, spec_raw_pe)
print("Saving the model")
+ start = time_ns()
model.save("model.h5")
+ t += [time_ns() - start]
+ t_names += ["Save"]
print("Loading the model")
+ start = time_ns()
model = Model()
model.load("model.h5")
+ t += [time_ns() - start]
+ t_names += ["Load"]
# test
print("Predict")
+ start = time_ns()
spec_pred = model.predict(pes_raw)
+ t += [time_ns() - start]
+ t_names += ["Predict"]
+
+ print("Time taken in ms")
+ df_time = pd.DataFrame(data=dict(time=t, name=t_names))
+ df_time.time *= 1e-6
+ print(df_time)
print("Plotting")
# plot
--
GitLab