--- a
+++ b/experiments/collect_bayes_stats.py
@@ -0,0 +1,26 @@
+from itertools import combinations
+
+import numpy as np
+import torch
+
+from models.segmentation_models import *
+
+
+def get_avg_dist_hist(model1, model2, filenames):
+    distance_matrix = np.zeros((len(filenames), len(filenames)))
+    with torch.no_grad():
+        for idx_pair in combinations(range(len(filenames)), 2):
+            model1.load_state_dict(torch.load(filenames[idx_pair[0]]))
+            model2.load_state_dict(torch.load(filenames[idx_pair[1]]))
+            dists = []
+            for (param1, param2) in zip(model1.parameters(), model2.parameters()):
+                dists.append(np.abs(torch.sum((param1 - param2))))
+            print(np.mean(dists))
+
+
+if __name__ == '__main__':
+    get_avg_dist_hist(DeepLab(), DeepLab(),
+                      ["Predictors/Augmented/DeepLab/pretrainmode=imagenet_{}".format(i) for i in range(3)])
+    print("vanilla:")
+    get_avg_dist_hist(DeepLab(), DeepLab(),
+                      ["Predictors/Vanilla/DeepLab/pretrainmode=imagenet_250_epochs_{}".format(i) for i in range(4)])