From eed28bdc3056a21bdc9f4ac1fabb51dd628ad265 Mon Sep 17 00:00:00 2001
From: Danilo Ferreira de Lima <danilo.enoque.ferreira.de.lima@xfel.de>
Date: Tue, 10 Jan 2023 16:18:07 +0100
Subject: [PATCH] Allow for no smoothing.
---
pes_to_spec/model.py | 17 +++++++++++++----
1 file changed, 13 insertions(+), 4 deletions(-)
diff --git a/pes_to_spec/model.py b/pes_to_spec/model.py
index ffc01bb..7e49e89 100644
--- a/pes_to_spec/model.py
+++ b/pes_to_spec/model.py
@@ -8,7 +8,6 @@ from autograd import grad
from scipy.signal import fftconvolve
from scipy.signal import find_peaks_cwt
from scipy.optimize import fmin_l_bfgs_b
-from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import KernelPCA, PCA
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.base import TransformerMixin, BaseEstimator
@@ -34,6 +33,7 @@ class PromptNotFoundError(Exception):
def __str__(self) -> str:
return "No prompt peak has been detected."
+
class HighResolutionSmoother(TransformerMixin, BaseEstimator):
"""
Smoothens out the high resolution data.
@@ -72,6 +72,8 @@ class HighResolutionSmoother(TransformerMixin, BaseEstimator):
Returns: Smoothened out spectrum.
"""
+ if self.high_res_sigma <= 0:
+ return X
# use a default energy axis is none is given
# assume only the energy step
energy = np.broadcast_to(self.energy, X.shape)
@@ -449,6 +451,8 @@ class Model(TransformerMixin, BaseEstimator):
Set to None to perform no selection.
validation_size: Fraction (number between 0 and 1) of the data to take for
validation and systematic uncertainty estimate.
+ n_pca_nonlinear: Number of nonlinear PCA components added at the preprocessing stage
+ to obtain nonlinearities as an input and improve the prediction.
"""
def __init__(self,
@@ -465,9 +469,13 @@ class Model(TransformerMixin, BaseEstimator):
if n_pca_nonlinear <= 0:
x_pca_model = PCA(n_pca_lr, whiten=True)
else:
- x_pca_model = FeatureUnion([('linear', PCA(n_pca_lr, whiten=True)),
+ x_pca_model = FeatureUnion([
+ ('linear', PCA(n_pca_lr, whiten=True)),
('nonlinear', Pipeline([('prep', PCA(n_pca_lr, whiten=True)),
- ('kpca', KernelPCA(n_pca_nonlinear, kernel='rbf', n_jobs=-1)),
+ ('kpca', KernelPCA(n_pca_nonlinear,
+ kernel='rbf',
+ gamma=0.1,
+ n_jobs=-1)),
])),
])
self.x_model = Pipeline([
@@ -475,7 +483,8 @@ class Model(TransformerMixin, BaseEstimator):
('pca', x_pca_model),
('unc', UncertaintyHolder()),
])
- self.y_model = Pipeline([('smoothen', HighResolutionSmoother(high_res_sigma)),
+ self.y_model = Pipeline([
+ ('smoothen', HighResolutionSmoother(high_res_sigma)),
('pca', PCA(n_pca_hr, whiten=False)),
('unc', UncertaintyHolder()),
])
--
GitLab