log some metrics after PCA and KMeans (other algorithms are untested for now). closes #31 for now

utils/algo.py

@@ -4,6 +4,7 @@ import tempfile
 import numpy as np
 import inspect
 import joblib
+from collections import Counter
 from pathlib import Path
 from typing import Dict, Any
 from qgis.core import (Qgis,
@@ -29,6 +30,7 @@ import sklearn.decomposition as decomposition
 import sklearn.cluster as cluster
 from sklearn.base import BaseEstimator
 from sklearn.preprocessing import StandardScaler
+from sklearn.metrics import silhouette_score, silhouette_samples
 
 if __name__ != "__main__":
     from .misc import get_unique_filename, calculate_chunk_size
@@ -632,6 +634,8 @@ class SKAlgorithm(IAMAPAlgorithm):
             iter = get_iter(model, fit_raster)
             model = self.fit_model(model, fit_raster, iter, feedback)
 
+        self.print_transform_metrics(model, feedback)
+        self.print_cluster_metrics(model,fit_raster, feedback)
         feedback.pushInfo(f'Fitting done, saving model\n')
         save_file = f'{self.method_name}.pkl'.lower()
         if self.save_model:
@@ -873,6 +877,37 @@ class SKAlgorithm(IAMAPAlgorithm):
 
         return help_str
 
+    def print_transform_metrics(self, model, feedback):
+        """
+        Log common metrics after a PCA.
+        """
+        
+        if hasattr(model, 'explained_variance_ratio_'):
+            # Explained variance ratio
+            explained_variance_ratio = model.explained_variance_ratio_
+
+            # Cumulative explained variance
+            cumulative_variance = np.cumsum(explained_variance_ratio)
+
+            # Loadings (Principal axes)
+            loadings = model.components_.T * np.sqrt(model.explained_variance_)
+
+            feedback.pushInfo(f'Explained Variance Ratio : \n{explained_variance_ratio}')
+            feedback.pushInfo(f'Cumulative Explained Variance : \n{cumulative_variance}')
+            feedback.pushInfo(f'Loadings (Principal axes) : \n{loadings}')
+
+    def print_cluster_metrics(self, model, fit_raster ,feedback):
+        """
+        Log common metrics after a Kmeans.
+        """
+        
+        if hasattr(model, 'inertia_'):
+
+            feedback.pushInfo(f'Inertia : \n{model.inertia_}')
+            feedback.pushInfo(f'Cluster sizes : \n{Counter(model.labels_)}')
+            ## silouhette score seem to heavy for now
+            # feedback.pushInfo(f'Silhouette Score : \n{silhouette_score(fit_raster, model.labels_)}')
+            # feedback.pushInfo(f'Silouhette Values : \n{silhouette_values(fit_raster, model.labels_)}')
 
     # used to handle any thread-sensitive cleanup which is required by the algorithm.
     def postProcessAlgorithm(self, context, feedback) -> Dict[str, Any]: