Bugfixes

net2brain/evaluations/encoding.py

@@ -684,8 +684,7 @@ def _linear_encoding(feat_path,
                 r = np.mean(r_lst) if not veRSA else r_lst
 
                 # Store correlation results
-                if return_correlations:
-                    corr_dict[fold_ii][layer_id][roi_name] = r_lst
+                corr_dict[fold_ii][layer_id][roi_name] = r_lst
 
                 all_rois_dict[fold_ii][layer_id][roi_name] = r
 
@@ -712,16 +711,18 @@ def _linear_encoding(feat_path,
                 # Compute the Standard Error of the Mean (SEM)
                 sem_value = sem(r_values_across_folds)
 
-            # If there is only one fold, use the r_lst from the fold directly for testing
+            # If there is only one fold, use the correlations from the fold directly for testing
             else:
                 # Get R Value
                 R = all_rois_dict[0][layer_id][roi_name]
 
                 if not veRSA:
+                    # Retrieve the per-voxel correlations for this specific layer/ROI
+                    r_lst_this = corr_dict[0][layer_id][roi_name]
                     # Perform t-test on the r_lst values since they are also correlation values
-                    _, significance = ttest_1samp(r_lst, 0)
+                    _, significance = ttest_1samp(r_lst_this, 0)
                     # Compute the Standard Error of the Mean (SEM)
-                    sem_value = sem(r_lst)
+                    sem_value = sem(r_lst_this)
                 else:
                     significance = None
                     sem_value = None
@@ -799,7 +800,7 @@ def train_Ridgeregression_per_ROI(trn_x, tst_x, trn_y, tst_y, veRSA=False, save_
 
         if save_model:
             with open(save_path, "wb") as f:
-                pickle.dump(reg, f)
+                pickle.dump(best_model, f)
     else:
         with open(save_path, "rb") as f:
             reg = pickle.load(f)
@@ -935,8 +936,8 @@ def _ridge_encoding(feat_path,
                 r = np.mean(r_lst) if not veRSA else r_lst
 
                 # Store correlation results
-                if return_correlations:
-                    corr_dict[fold_ii][layer_id][roi_name] = r_lst
+                # if return_correlations:
+                corr_dict[fold_ii][layer_id][roi_name] = r_lst
 
                 all_rois_dict[fold_ii][layer_id][roi_name] = r
 
@@ -963,16 +964,18 @@ def _ridge_encoding(feat_path,
                 # Compute the Standard Error of the Mean (SEM)
                 sem_value = sem(r_values_across_folds)
 
-            # If there is only one fold, use the r_lst from the fold directly for testing
+            # If there is only one fold, use the correlations from the fold directly for testing
             else:
                 # Get R Value
                 R = all_rois_dict[0][layer_id][roi_name]
 
                 if not veRSA:
+                    # Retrieve the per-voxel correlations for this specific layer/ROI
+                    r_lst_this = corr_dict[0][layer_id][roi_name]
                     # Perform t-test on the r_lst values since they are also correlation values
-                    _, significance = ttest_1samp(r_lst, 0)
+                    _, significance = ttest_1samp(r_lst_this, 0)
                     # Compute the Standard Error of the Mean (SEM)
-                    sem_value = sem(r_lst)
+                    sem_value = sem(r_lst_this)
                 else:
                     significance = None
                     sem_value = None

net2brain/evaluations/rsa.py

@@ -360,17 +360,16 @@ def compare_model(self, other_RSA):
 
             r_name, _, r_array_name = self.get_rnames()
 
-            model_ii = np.argmin([layer_dict[r_name] for layer_dict in model_layers_dict])
-            other_ii = np.argmin([layer_dict[r_name] for layer_dict in other_layers_dict])
+            model_ii = np.argmax([layer_dict[r_name] for layer_dict in model_layers_dict])
+            other_ii = np.argmax([layer_dict[r_name] for layer_dict in other_layers_dict])
 
             tstat, p = stats.ttest_ind(other_layers_dict[other_ii][r_array_name][0], model_layers_dict[model_ii][r_array_name][0])
 
             scan_key = "(" + str(counter) + ") " + roi[:-4]
 
             comp_dic[scan_key] = (tstat, p)
-            if p < 0.5:
+            if p < 0.05:
                 sig_pair = sorted(((scan_key, self.model_name), (scan_key, other_RSA.model_name)),
                                   key=lambda element: (element[1]))
                 sig_pairs.append(sig_pair)
-        return comp_dic, sig_pairs
-
+        return comp_dic, sig_pairs

net2brain/evaluations/searchlight.py

@@ -283,12 +283,12 @@ def evaluate(self):
         # TODO: Add actual NoiseCeiling!
         noise_ceiling = {'lnc': 0, 'unc': 0}
 
-        for counter, layer in self.model_rdms:
+        for counter, layer in enumerate(self.model_rdms):
             self.current_layer = layer
             self.layer_counter = counter
 
-            this_model_rdm = [np.load(op.join(self.model_rdms_path, layer))['arr_0']]
-            this_model_rdm = self.check_squareform(this_model_rdm) # Check if rdm is squareform #TODO Remove soon after reimplementing RSA
+            this_model_rdm = np.load(op.join(self.model_rdms_path, layer))['arr_0']
+            this_model_rdm = [self.check_squareform(this_model_rdm)]
             self.evaluate_searchlight(noise_ceiling, this_model_rdm)
 
         return self.final_dict

net2brain/evaluations/variance_partitioning_analysis.py

@@ -616,6 +616,10 @@ def evaluate_4(self, dep_var, num_subjects, eeg_time):
         for i in range(3):
             for j in range(i+1, 4):
                 y_values[f'y{i+1}{j+1}'] = np.zeros((num_subjects, eeg_time))
+        for i in range(2):
+            for j in range(i+1, 3):
+                for k in range(j+1, 4):
+                    y_values[f'y{i+1}{j+1}{k+1}'] = np.zeros((num_subjects, eeg_time))
         y_values['y1234'] = np.zeros((num_subjects, eeg_time))
 
         R_un = np.zeros(eeg_time)
@@ -642,7 +646,11 @@ def evaluate_4(self, dep_var, num_subjects, eeg_time):
         # Define naming convention for the dataframe
         data_to_append = []
         for key, value in R_values.items():
-            desc = ' and '.join(key[1:]) + ' Influence' if len(key) > 2 else self.variable_names[int(key[1])-1] + ' Influence'
+            if len(key) <= 2:
+                desc = self.variable_names[int(key[1])-1] + ' Influence'
+            else:
+                desc = 'Combined ' + ' and '.join([self.variable_names[int(c)-1] for c in key[1:]]) + ' Influence'
+
             data_to_append.append({
                 'Variable': key,
                 'Description': desc,
@@ -651,7 +659,10 @@ def evaluate_4(self, dep_var, num_subjects, eeg_time):
                 'Color': None
             })
         for key, value in y_values.items():
-            desc = 'Shared Variance of ' + ' and '.join(key[1:]) if len(key) > 4 else 'Unique Variance ' + self.variable_names[int(key[1])-1]
+            if len(key) <= 2:
+                desc = 'Unique ' + self.variable_names[int(key[1])-1]
+            else:
+                desc = 'Shared Variance ' + ' and '.join([self.variable_names[int(c)-1] for c in key[1:]])
             data_to_append.append({
                 'Variable': key,
                 'Description': desc,

setup.py

@@ -2,7 +2,7 @@
 
 setup(
     name='net2brain',
-    version='1.2.6',
+    version='1.3',
     author='Roig Lab',
     packages=find_packages(),
     package_data={