Allow for no smoothing.

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()),
                                 ])