add largeZ model in comparison, but comment it out for now

ivadomed · Nov 11, 2024 · 1e21f2c · 1e21f2c
1 parent c5974b0
commit 1e21f2c
Showing 1 changed file with 22 additions and 6 deletions.
diff --git a/testing/generate_figures_and_compute_tests.py b/testing/generate_figures_and_compute_tests.py
@@ -23,6 +23,11 @@
      "Dataset904_tumMSStitchedStraightRegion": 'Stitched\nStraightened',
     #  "Dataset910_tumMSChunksPolyNYUAxialRegion": 'Chunks\nPolyNYU',
 }
+order_datasets_tum_largeZ = {
+     "Dataset902_tumMSStitchedRegion": 'Stitched\nNative', 
+     "Dataset904_tumMSStitchedStraightRegion": 'Stitched\nStraightened',
+    #  "Dataset910_tumMSChunksPolyNYUAxialRegion": 'Chunks\nPolyNYU',
+}
 
 order_datasets_testing_large = {
     "DeepSegLesionInference_tumNeuropoly": 'DeepSeg\nLesion',
@@ -85,9 +90,14 @@ def find_dataset_in_path(path):
 
 
 def find_model_in_path(path):
-    match = re.search(r'2d|3d_fullres', path)
+    match = re.search(r'2d|3d_fullres', path) #|3d_largeZ', path)
     if match:
-        return '2D' if '2d' in match.group(0) else '3D'
+        if '2d' in match.group(0):
+            return '2D'
+        elif '3d_fullres' in match.group(0):
+            return '3D'
+        elif '3d_largeZ' in match.group(0):
+            return '3D_largeZ'
 
 
 def find_filename_in_path(path):
@@ -131,7 +141,7 @@ def create_rainplot(args, df, metrics, path_figures, pred_type):
                           box_showmeans=True,  # show mean value inside the boxplots
                           box_meanprops={'marker': '^', 'markerfacecolor': 'black', 'markeredgecolor': 'black',
                                          'markersize': '6'},
-                          hue_order=['3D', '2D'],
+                          hue_order=['3D', '2D'] #, '3D_largeZ'],
                           )
 
         # TODO: include mean +- std for each boxplot above the mean value
@@ -149,7 +159,8 @@ def create_rainplot(args, df, metrics, path_figures, pred_type):
             labels[i] = f'{label} Model' + ' ($\it{n}$' + f' = {n})'
         # Since the figure contains violionplot + boxplot + scatterplot we are keeping only last two legend entries
         handles, labels = handles[-2:], labels[-2:]
-        ax.legend(handles, labels, fontsize=TICK_FONT_SIZE, loc='lower center', bbox_to_anchor=(0.5, -0.25), ncol=2)
+        ax.legend(handles, labels, fontsize=TICK_FONT_SIZE, loc='lower center', 
+                  bbox_to_anchor=(0.5, -0.25), ncol=len(labels))
 
         # Make legend box's frame color black and remove transparency
         legend = ax.get_legend()
@@ -372,12 +383,14 @@ def compute_kruskal_wallis_test_across_tum_datasets(df_concat, list_of_metrics):
 
         df_stitch_native_2d = df_concat[(df_concat['dataset'] == 'Dataset902_tumMSStitchedRegion') & (df_concat['model'] == '2D')]
         df_stitch_native_3d = df_concat[(df_concat['dataset'] == 'Dataset902_tumMSStitchedRegion') & (df_concat['model'] == '3D')]
+        # df_stitch_native_3d_largeZ = df_concat[(df_concat['dataset'] == 'Dataset902_tumMSStitchedRegion') & (df_concat['model'] == '3D_largeZ')]
 
         df_chunks_straight_2d = df_concat[(df_concat['dataset'] == 'Dataset903_tumMSChunksStraightRegion') & (df_concat['model'] == '2D')]
         df_chunks_straight_3d = df_concat[(df_concat['dataset'] == 'Dataset903_tumMSChunksStraightRegion') & (df_concat['model'] == '3D')]
 
         df_stitch_straight_2d = df_concat[(df_concat['dataset'] == 'Dataset904_tumMSStitchedStraightRegion') & (df_concat['model'] == '2D')]
         df_stitch_straight_3d = df_concat[(df_concat['dataset'] == 'Dataset904_tumMSStitchedStraightRegion') & (df_concat['model'] == '3D')]
+        # df_stitch_straight_3d_largeZ = df_concat[(df_concat['dataset'] == 'Dataset904_tumMSStitchedStraightRegion') & (df_concat['model'] == '3D_largeZ')]
 
         # ensure that the two dataframes have the same number of rows
         assert len(df_chunks_native_2d) == len(df_stitch_native_2d) == len(df_chunks_straight_2d) == len(df_stitch_straight_2d) == \
@@ -392,10 +405,12 @@ def compute_kruskal_wallis_test_across_tum_datasets(df_concat, list_of_metrics):
             f'{metric}_chunks_native_3d': df_chunks_native_3d[metric].values,
             f'{metric}_stitch_native_2d': df_stitch_native_2d[metric].values,
             f'{metric}_stitch_native_3d': df_stitch_native_3d[metric].values,
+            # f'{metric}_stitch_native_3d_largeZ': df_stitch_native_3d_largeZ[metric].values,
             f'{metric}_chunks_straight_2d': df_chunks_straight_2d[metric].values,
             f'{metric}_chunks_straight_3d': df_chunks_straight_3d[metric].values,
             f'{metric}_stitch_straight_2d': df_stitch_straight_2d[metric].values,
             f'{metric}_stitch_straight_3d': df_stitch_straight_3d[metric].values,
+            # f'{metric}_stitch_straight_3d_largeZ': df_stitch_straight_3d_largeZ[metric].values
         })
 
         # Print number of subjects
@@ -404,9 +419,10 @@ def compute_kruskal_wallis_test_across_tum_datasets(df_concat, list_of_metrics):
         # Compute Kruskal-Wallis H-test
         stat, p = kruskal(
             df[metric + '_chunks_native_2d'], df[metric + '_chunks_native_3d'],
-            df[metric + '_stitch_native_2d'], df[metric + '_stitch_native_3d'],
+            df[metric + '_stitch_native_2d'], df[metric + '_stitch_native_3d'], #df[metric + '_stitch_native_3d_largeZ'],
             df[metric + '_chunks_straight_2d'], df[metric + '_chunks_straight_3d'],
-            df[metric + '_stitch_straight_2d'], df[metric + '_stitch_straight_3d'])
+            df[metric + '_stitch_straight_2d'], df[metric + '_stitch_straight_3d'], #df[metric + '_stitch_straight_3d_largeZ']
+            )
         logger.info(f'{metrics_short[metric]}: Kruskal-Wallis H-test: formatted p{format_pvalue(p)}, unformatted p={p:0.6f}')
 
         if p < 0.05: