separate functions for reading and creating metric table files

scripts/plot_hyperparams.py

@@ -159,8 +159,16 @@ def _plot_scatters(data_dir: pathlib.Path, metric: str) -> plt.Figure:
     """
     data_dirs = list(data_dir.glob("*"))
     runs = []
+
+    # Create metrics files for all the runs
+    # If the run is still going, there might be some missing files
     for dir_ in tqdm(data_dirs, desc="Creating metric tables"):
-        # We might still be running, in which case the last dir will be incomplete
+        try:
+            _write_metrics_files(dir_)
+        except FileNotFoundError:
+            continue
+
+    for dir_ in data_dirs:
         if metric == "dice":
             try:
                 score = _dicescore(dir_)
@@ -250,38 +258,42 @@ def _plot_med(input_dir: pathlib.Path, output_dir: pathlib.Path):
     fig.savefig(output_dir / "scores.png")
 
 
-def _dicescore(results_dir: pathlib.Path) -> float:
+def _write_metrics_files(results_dir: pathlib.Path) -> None:
     """
-    Get the DICE score from the i-th run
+    Write files containing the validation metrics for each run, as markdown
+
+    :param results_dir: The directory containing the results from a run
 
     """
     # We'll write the score to a file
     metrics_file = results_dir / "metrics.txt"
 
-    # If it doesn't exist, write the metrics file
-    if not metrics_file.exists():
-        n_val_imgs = len(list(results_dir.glob("*val_pred_*.npy")))
-        assert n_val_imgs == len(list(results_dir.glob("val_truth_*.npy")))
-
-        if n_val_imgs == 0:
-            raise FileNotFoundError(f"No validation images found in {results_dir}")
+    if metrics_file.exists():
+        return
 
-        pred = [np.load(results_dir / f"val_pred_{i}.npy") for i in range(n_val_imgs)]
-        truth = [np.load(results_dir / f"val_truth_{i}.npy") for i in range(n_val_imgs)]
+    n_val_imgs = len(list(results_dir.glob("*val_pred_*.npy")))
+    assert n_val_imgs == len(list(results_dir.glob("val_truth_*.npy")))
+    if n_val_imgs == 0:
+        raise FileNotFoundError(f"No validation images found in {results_dir}")
 
-        for p, t in zip(pred, truth):
-            assert p.shape == t.shape, f"{p.shape=} != {t.shape=}"
+    # Load the arrays
+    pred = [np.load(results_dir / f"val_pred_{i}.npy") for i in range(n_val_imgs)]
+    truth = [np.load(results_dir / f"val_truth_{i}.npy") for i in range(n_val_imgs)]
+    for p, t in zip(pred, truth):
+        assert p.shape == t.shape, f"{p.shape=} != {t.shape=}"
+        if not p.min() >= 0 and p.max() <= 1:
+            raise ValueError("Prediction should be scaled to between 0 and 1")
 
-            # The prediction should already be scaled to be between 0 and 1
-            if not p.min() >= 0 and p.max() <= 1:
-                raise ValueError("Prediction should be scaled to between 0 and 1")
+    table = metrics.table(truth, pred)
+    with open(metrics_file, "w", encoding="utf-8") as f:
+        f.write(table.to_markdown())
 
-        table = metrics.table(truth, pred)
 
-        with open(metrics_file, "w", encoding="utf-8") as f:
-            f.write(table.to_markdown())
+def _metrics_df(metrics_file: pathlib.Path) -> pd.DataFrame:
+    """
+    Read the metrics files and return a DataFrame
 
-    # get the average DICE score from the metrics file
+    """
     df = (
         pd.read_table(
             metrics_file,
@@ -295,6 +307,24 @@ def _dicescore(results_dir: pathlib.Path) -> float:
         .astype(float)
     )
     df.columns = df.columns.str.strip()
+    return df
+
+
+def _dicescore(results_dir: pathlib.Path) -> float:
+    """
+    Get the DICE score from the i-th run
+
+    :param results_dir: The directory containing the results from a run
+
+    """
+    # We'll write the score to a file
+    metrics_file = results_dir / "metrics.txt"
+
+    if not metrics_file.exists():
+        raise FileNotFoundError(f"No metrics file found in {results_dir}")
+
+    # get the average DICE score from the metrics file
+    df = _metrics_df(metrics_file)
 
     return df["Dice"].mean()