update label outliers script

mu_map/random_search/eval/label_outliers.py

@@ -27,7 +27,7 @@ Controls:
 """
 
 default_rs_dir = "cgan_random_search/"
-default_outfile = os.path.join(default_rs_dir, "outliers.csv")
+default_outfile = "outliers.csv"
 
 parser = argparse.ArgumentParser(
     description="label random search runs as outliers (not-converged) by having a look at the model's prediction on the validation split of the dataset",
@@ -53,6 +53,12 @@ parser.add_argument(
 )
 args = parser.parse_args()
 
+args.out = os.path.join(args.random_search_dir, args.out)
+
+wname = "Label Outlier"
+cv.namedWindow(wname, cv.WINDOW_NORMAL)
+cv.resizeWindow(wname, 1600, 900)
+
 device = torch.device(args.device)
 runs = load_data(args.random_search_dir)
 
@@ -66,7 +72,8 @@ print(controls)
 
 total = str(len(runs))
 for i, run in enumerate(runs):
-    print(f"Run {str(i+1):>{len(total)}}/{total}", end="\r")
+    nmae = runs[run]["measures"]["NMAE"].mean()
+    print(f"Run {str(i+1):>{len(total)}}/{total} - with NMAE {nmae:.5f}", end="\r")
 
     if run in data["run"]:
         continue
@@ -94,13 +101,11 @@ for i, run in enumerate(runs):
     model.load_state_dict(weights)
     model = model.to(device).eval()
 
-    wname = "Label Outlier"
-    cv.namedWindow(wname, cv.WINDOW_NORMAL)
-    cv.resizeWindow(wname, 1600, 900)
-
     def action(key):
         if key == ord("o"):
-            print(f"Run {str(i+1):>{len(total)}}/{total} - Outlier!")
+            print(
+                f"Run {str(i+1):>{len(total)}}/{total} - with NMAE {nmae:.5f} - Outlier!"
+            )
             data["outlier"][i] = True
             return True