Fixing minor issues for figures display/closing [skip ci]

Author: dbouget

raidionicsval/Plotting/metric_versus_binned_boxplot.py

@@ -74,6 +74,7 @@ def compute_binned_metric_over_metric_boxplot(folder, data, metric1, metric2, cr
         sns_plot = sns.boxplot(x=data[metric1], y=binned)
         sns_plot.figure.savefig(os.path.join(folder, 'boxplot_' + metric1 + '_over_' + metric2 + '_normal_bins' + postfix + '.png'),
                                 dpi=300, bbox_inches="tight")
+        plt.close(sns_plot.figure)
 
         binned, edges = pd.qcut(data[metric2], q=nb_bins, retbins=True, precision=2, duplicates='drop')
         nb_bins = len(edges) - 1
@@ -125,6 +126,7 @@ def compute_binned_metric_over_metric_boxplot(folder, data, metric1, metric2, cr
         sns_plot.figure.savefig(os.path.join(folder, 'boxplot_' + metric1 + '_over_' + metric2 + '_equal_bins' + postfix + '.png'),
                                 dpi=300, bbox_inches="tight")
         # plt.show()
+        plt.close(sns_plot.figure)
 
         print('\n\nTotal selected: {}, total to find: {}. Recall: {:.4f}'.format(total_selected, total_to_find,
                                                                                  total_selected / total_to_find))

raidionicsval/Studies/AbstractStudy.py

@@ -271,6 +271,9 @@ def compute_and_plot_metric_over_metric_categories(self, class_name: str, data=N
             for cat in optimal_results_per_cutoff.keys():
                 # @TODO. Must include a new fold average specific for the studies, with mean and std values as input,
                 # which is different from the inputs to the computation in the validation part....
+                if len(optimal_results_per_cutoff[cat]) == 0:
+                    print("Skipping analysis for {} {}. Collected pd.DataFrame is empty.\n".format(metric2, cat))
+                    return
                 self.compute_fold_average(self.output_folder, data=optimal_results_per_cutoff[cat],
                                           class_optimal=self.classes_optimal, metrics=self.metric_names,
                                           suffix=suffix + '_' + metric2 + '_' + cat,

raidionicsval/Validation/extra_metrics_computation.py

@@ -310,8 +310,26 @@ def compute_overall_metrics_correlation(input_folder, output_folder, data=None,
     results_for_matrix_df = results_for_matrix_df.dropna()
     results_for_matrix_df = results_for_matrix_df.apply(pd.to_numeric)
     corr_matrix = results_for_matrix_df.corr()
-    # print(corr_matrix.style.background_gradient(cmap='coolwarm').set_precision(2).render())
     export_correlation_matrix_to_latex(output_folder, corr_matrix, suffix='_' + class_name + suffix)
+    export_correlation_matrix_to_html(output_folder, corr_matrix, suffix='_' + class_name + suffix)
+
+
+def export_correlation_matrix_to_html(output_folder, matrix, suffix=""):
+    """
+    Converts and saves the correlation matrix as an html table
+
+    ...
+    Attributes
+    ----------
+    output_folder: str
+        Destination folder where the latex table will be saved under correlation_matrix.html
+    matrix: pd.DataFrame
+        Correlation matrix object as a pandas DataFrame
+    suffix: (optional) str
+        Name to append to the end of the destination file
+    """
+    matrix_filename = os.path.join(output_folder, 'correlation_matrix' + suffix + '.html')
+    matrix.style.background_gradient(cmap='coolwarm').to_html(matrix_filename)
 
 
 def export_correlation_matrix_to_latex(output_folder, matrix, suffix=""):

raidionicsval/Validation/validation_utilities.py

@@ -144,6 +144,10 @@ def best_segmentation_probability_threshold_analysis_inner(folder, detection_ove
     fig4.savefig(os.path.join(folder, 'Validation', 'OptimalSearch', 'metrics_scatter_over_threshold.png'), dpi=300,
                  bbox_inches="tight")
 
+    plt.close(fig)
+    plt.close(fig2)
+    plt.close(fig3)
+    plt.close(fig4)
     return max_overlap, max_threshold