Merge branch '5-add-plots-2' into 'wip_clean_scripts'

Resolve "Add plots"

See merge request !5

.gitignore

@@ -370,3 +370,6 @@ TSWLatexianTemp*
 *.lpz
 
 reports/*.pdf
+
+# Image
+*.png

code/bolsonaro/trainer.py

@@ -30,6 +30,8 @@ class Trainer(object):
             y_omp = y_forest
             self._logger.debug('Fitting both the forest and OMP on train+dev subsets.')
 
+        # TODO: add an option to train forest to train+dev and OMP to dev
+
         model.fit(
             X_forest=X_forest,
             y_forest=y_forest,

code/bolsonaro/visualization/plotter.py

 import matplotlib.pyplot as plt
 import numpy as np
-from sklearn.neighbors.kde import KernelDensity
+import pandas as pd
 
 
 class Plotter(object):
 
     @staticmethod
-    def weight_density(weights, X, file_path):
-        X_plot = [np.exp(elem) for elem in weights]
-        fig, ax = plt.subplots()
+    def weight_density(all_experiment_weights, file_path):
+        '''
+        Function that creates the figure with the density of the weights
+        :param all_experiment_weights: The weights for the different experiments
+        :param file path: str, path where the figure will be saved
+        '''
 
-        for kernel in ['gaussian', 'tophat', 'epanechnikov']:
-            kde = KernelDensity(kernel=kernel, bandwidth=0.5).fit(X_plot)
-            log_dens = kde.score_samples(X_plot)
-            ax.plot(X_plot, np.exp(log_dens), '-',
-                    label="kernel = '{0}'".format(kernel))
+        all_experiment_weights = np.array(list(all_experiment_weights.values()))
+        n = len(all_experiment_weights)
+        colors = Plotter.get_colors_from_cmap(n)
 
-        ax.legend(loc='upper left')
-        ax.plot(X[:, 0], -0.005 - 0.01 * np.random.random(X.shape[0]), '+k')
+        fig, ax = plt.subplots()
+        for i in range(n):
+            for weights in all_experiment_weights[i]:
+                pd.Series([weight for weight in weights if weight != 0]).plot.kde(
+                    figsize=(15, 10), ax=ax, color=colors[i])
 
-        ax.set_xlim(-4, 9)
-        ax.set_ylim(-0.02, 0.4)
+        ax.set_title('Density weights of the OMP')
         fig.savefig(file_path, dpi=fig.dpi)
         plt.close(fig)
 

code/compute_results.py

@@ -59,6 +59,8 @@ if __name__ == "__main__":
         experiment_dev_scores = dict()
         experiment_test_scores = dict()
 
+        experiment_weights = dict()
+
         # Used to check if all losses were computed using the same metric (it should be the case)
         experiment_score_metrics = list()
 
@@ -74,6 +76,8 @@ if __name__ == "__main__":
             experiment_dev_scores[seed] = list()
             experiment_test_scores[seed] = list()
 
+            experiment_weights[seed] = list()
+
             # List the forest sizes in models/{experiment_id}/seeds/{seed}/extracted_forest_size
             extracted_forest_sizes = os.listdir(extracted_forest_size_root_path)
             for extracted_forest_size in extracted_forest_sizes:
@@ -84,9 +88,13 @@ if __name__ == "__main__":
                 # Load [...]/model_parameters.json file and build the model using these parameters and the weights and forest from model_raw_results.pickle
                 model = ModelFactory.load(dataset.task, extracted_forest_size_path, experiment_id, model_raw_results)
                 # Save temporarly some raw results (TODO: to complete to retreive more results)
+                # Save the scores
                 experiment_train_scores[seed].append(model_raw_results.train_score)
                 experiment_dev_scores[seed].append(model_raw_results.dev_score)
                 experiment_test_scores[seed].append(model_raw_results.test_score)
+                # Save the weights
+                experiment_weights[seed].append(model_raw_results.weights)
+                # Save the metric
                 experiment_score_metrics.append(model_raw_results.score_metric)
 
         if len(set(experiment_score_metrics)) > 1:
@@ -107,3 +115,9 @@ if __name__ == "__main__":
             all_labels=['train', 'dev', 'test'],
             title='Loss values of the trained model'
         )
+
+        # Plot the density of the weights
+        Plotter.weight_density(
+            file_path=args.results_dir + os.sep + experiment_id + os.sep + 'density_weight.png',
+            all_experiment_weights=experiment_weights
+        )

requirements.txt

@@ -11,4 +11,5 @@ python-dotenv>=0.5.1
 scikit-learn
 python-dotenv
 tqdm
-matplotlib
+matplotlib
\ No newline at end of file
+pandas
\ No newline at end of file