Switched to scipy's kde, which is smart enough to factorize the mean.

pes_to_spec/__init__.py

@@ -2,4 +2,4 @@
 Estimate high-resolution photon spectrometer data from low-resolution non-invasive measurements.
 """
 
-VERSION = "0.1.2"
+VERSION = "0.1.3"

pes_to_spec/model.py

@@ -13,7 +13,7 @@ from sklearn.pipeline import Pipeline
 from sklearn.kernel_approximation import Nystroem
 #from sklearn.linear_model import ARDRegression
 from sklearn.linear_model import BayesianRidge
-from sklearn.neighbors import KernelDensity
+from scipy.stats import gaussian_kde
 from sklearn.ensemble import IsolationForest
 from functools import reduce
 from itertools import product
@@ -765,14 +765,13 @@ class Model(TransformerMixin, BaseEstimator):
 
         Return: weights.
         """
-        self.kde_xgm = KernelDensity(bandwidth="scott", kernel="gaussian")
-        self.kde_xgm.fit(intensity.reshape(-1, 1))
+        self.kde_xgm = gaussian_kde(intensity.reshape(-1), bw_method="scott")
         self.mu_xgm = np.mean(intensity.reshape(-1), axis=0)
         self.sigma_xgm = np.std(intensity.reshape(-1), axis=0)
         q = np.quantile(intensity, [0.10, 0.90])
         l, h = q[0], q[1]
         x = intensity*((intensity > l) & (intensity < h)) + l*(intensity <= l) + h*(intensity >= h)
-        log_prob = self.kde_xgm.score_samples(x.reshape(-1, 1))
+        log_prob = self.kde_xgm.logpdf(x.reshape(-1))
         w = np.exp(-log_prob)
         w = w/np.median(w)
         return w
@@ -920,8 +919,7 @@ class Model(TransformerMixin, BaseEstimator):
 
         # get intensity effect
         intensity = np.sum(z, axis=1)
-        self.kde_intensity = KernelDensity(bandwidth="scott", kernel="gaussian")
-        self.kde_intensity.fit(intensity.reshape(-1, 1))
+        self.kde_intensity = gaussian_kde(intensity.reshape(-1), bw_method="scott")
         self.mu_intensity = np.mean(intensity.reshape(-1), axis=0)
         self.sigma_intensity = np.std(intensity.reshape(-1), axis=0)
 
@@ -970,11 +968,11 @@ class Model(TransformerMixin, BaseEstimator):
         low_pca = pca_model.transform(low_res)
         return self.ood['full'].predict(low_pca)
 
-    def xgm_profile(self) -> KernelDensity:
+    def xgm_profile(self) -> gaussian_kde:
         """Get KDE for the XGM intensity."""
         return self.kde_xgm
 
-    def intensity_profile(self) -> KernelDensity:
+    def intensity_profile(self) -> gaussian_kde:
         """Get KDE for the predicted intensity."""
         return self.kde_intensity