rerun bootstrapping, add some more ablations, gather final tables

.github/CODEOWNERS

@@ -82,6 +82,7 @@ scripts/4-z-* @StevenSong
 scripts/5-1-run-ecgfounder-logreg.sh @StevenSong
 scripts/5-2-run-stmem-logreg.sh @StevenSong
 scripts/6-run-protossl-heedb-pia.sh @StevenSong
+scripts/6-z-* @StevenSong
 scripts/7-run-protossl-heedb-pit.sh @StevenSong
 scripts/8-run-protossl-heedb-pip.sh @StevenSong
 scripts/9-0-run-ecgfounder-patches.sh @StevenSong

protossl/datasets/_cinc_dataset.py

@@ -40,6 +40,7 @@ def __init__(
         _path = Path(dataset_path)
         df = pd.read_csv(_path / "georgia.csv")
         df = df[df["split"] == split]
+        self._df = df.reset_index(drop=True)
 
         self.source_ids = torch.as_tensor(df["patient_id"].to_numpy())
         self.sample_ids = torch.as_tensor(df["ecg_id"].to_numpy())

protossl/datasets/_code15_dataset.py

@@ -40,6 +40,7 @@ def __init__(
         _path = Path(dataset_path)
         df = pd.read_csv(_path / "labels.csv")
         df = df[df["split"] == split]
+        self._df = df.reset_index(drop=True)
 
         self.source_ids = torch.as_tensor(df["patient_id"].to_numpy())
         self.sample_ids = torch.as_tensor(df["exam_id"].to_numpy())

protossl/datasets/_echonext_dataset.py

@@ -41,6 +41,7 @@ def __init__(
         df = pd.read_csv(_path / "EchoNext_metadata_100k.csv")
         df = df.rename(columns=mapping)
         split_mask = df["split"] == split
+        self._df = df.loc[split_mask].reset_index(drop=True)
         id_df = df.loc[split_mask, ["patient_key", "ecg_key"]].reset_index(drop=True)
         label_df = df.loc[split_mask, target_cols].reset_index(drop=True)
 

protossl/datasets/_mimic_dataset.py

@@ -43,6 +43,7 @@ def __init__(
         _path = Path(dataset_path)
         df = pd.read_csv(_path / "ed-ecgs.csv")
         df = df[df["split"] == split]
+        self._df = df.reset_index(drop=True)
 
         self.source_ids = torch.as_tensor(df["subject_id"].to_numpy())
         self.sample_ids = torch.as_tensor(df["study_id"].to_numpy())

protossl/datasets/_ptbxl_dataset.py

@@ -66,6 +66,7 @@ def __init__(
         else:
             raise ValueError(f"Unknown split: {split}")
         df = df[mask]
+        self._df = df.reset_index(drop=True)
 
         self.source_ids = torch.as_tensor(df["patient_id"].astype(int).to_numpy())
         self.sample_ids = torch.as_tensor(df.index.to_numpy())

protossl/datasets/_zzu_dataset.py

@@ -57,6 +57,7 @@ def __init__(
             )
         df = get_zzu_dataframe(dataset_path)
         df = df[df["split"] == split]
+        self._df = df.reset_index(drop=True)
 
         self.source_ids = torch.as_tensor(df["Patient_ID"].to_numpy())
         self.sample_ids = torch.as_tensor(df["ECG_ID"].to_numpy())

results/audio-results.ipynb

@@ -52,15 +52,15 @@
     "            his = []\n",
     "            for i in range(n_folds):\n",
     "                fold_path = Path(fpath.format(i))\n",
-    "                metrics = pd.read_csv(fold_path / model_dir / \"metrics-bootstrapped.csv\", index_col=\"Label\")\n",
+    "                metrics = pd.read_csv(fold_path / model_dir / \"metrics-bootstrapped-v2.csv\", index_col=\"Label\")\n",
     "                accs.append(metrics.loc[\"Multiclass\", \"Accuracy\"])\n",
     "                los.append(metrics.loc[\"Multiclass\", \"Accuracy 95% CI (lo)\"])\n",
     "                his.append(metrics.loc[\"Multiclass\", \"Accuracy 95% CI (hi)\"])\n",
     "            acc = np.mean(accs)\n",
     "            lo = np.mean(los)\n",
     "            hi = np.mean(his)\n",
     "        else:\n",
-    "            metrics = pd.read_csv(Path(fpath) / model_dir / \"metrics-bootstrapped.csv\", index_col=\"Label\")\n",
+    "            metrics = pd.read_csv(Path(fpath) / model_dir / \"metrics-bootstrapped-v2.csv\", index_col=\"Label\")\n",
     "            acc = metrics.loc[\"Multiclass\", \"Accuracy\"]\n",
     "            lo = metrics.loc[\"Multiclass\", \"Accuracy 95% CI (lo)\"]\n",
     "            hi = metrics.loc[\"Multiclass\", \"Accuracy 95% CI (hi)\"]\n",