Add normalized regression age

sage/trainer/utils.py

@@ -3,8 +3,10 @@
 from pathlib import Path
 import pickle
 
+import numpy as np
 import pandas as pd
 import matplotlib.pyplot as plt
+from sklearn.metrics import confusion_matrix
 import seaborn as sns
 sns.set_theme()
 import torch
@@ -91,7 +93,7 @@ def finalize_inference(prediction: list,
     preds, target = prediction["pred"], prediction["target"]
     if name.startswith("C"):
         logger.info("Classification data given:")
-        _cls_infrence(preds=preds, target=target, root_dir=root_dir, run_name=run_name)
+        _cls_inference(preds=preds, target=target, root_dir=root_dir, run_name=run_name)
     elif name[0] in set(["R", "M"]):
         logger.info("Regression data given:")
         _reg_inference(preds=preds, target=target, root_dir=root_dir, run_name=run_name)
@@ -135,8 +137,9 @@ def _get_norm_cf_reg(preds, target, root_dir, run_name) -> None:
     2. Interval of each bin: 5 or 10.
     """
     AGEMIN, AGEMAX = 0, 100
-    min_, max_ = target.min(), target.max()
-    min_, max_ = max(AGEMIN, min_), min(AGEMAX, max_)
+    pmin_, pmax_ = preds.min(), preds.max()
+    tmin_, tmax_ = target.min(), target.max()
+    min_, max_ = max(AGEMIN, min(tmin_, pmin_)), min(AGEMAX, max(tmax_, pmax_))
     int_ = max_ - min_ # interval
 
     # Check two intervals 5 and 10
@@ -149,14 +152,38 @@ def _get_norm_cf_reg(preds, target, root_dir, run_name) -> None:
     while (ub % bin_) != 0:
         ub += 1
     bins = [lb + bin_ * idx for idx in range(0, (ub - lb) // bin_ + 1)]
-    labels = [f"{left+1}-{right}" for left, right in zip(bins, bins[1:])]
+    labels = [f"{left}-{right - 1}" for left, right in zip(bins, bins[1:])]
+
+    cut_kwargs = dict(bins=bins, labels=labels, include_lowest=True)
+    _preds = pd.cut(x=preds, **cut_kwargs)
+    _target = pd.cut(x=target, **cut_kwargs)
+    assert (_preds.isna().sum() + _target.isna().sum()) == 0, f"nan value in binning."
+
+    fig, ax = plt.subplots(figsize=(13, 6), ncols=2)
+    labelsize = "large"
+    titlesize = "xx-large"
+
+    # Normalized Confusion Matrix
+    cf = confusion_matrix(_preds, _target)
+    sns.heatmap(cf, annot=True, fmt="d", xticklabels=_target.categories, ax=ax[0])
+    ax[0].set_xlabel("Target", size=labelsize)
+    ax[0].set_yticklabels(_preds.categories, rotation=270)
+    ax[0].set_ylabel("Prediction", size=labelsize)
+
+    # 0-1 row-wise Noramlized Confusion.
+    norm_cf = (cf.T / cf.sum(axis=1)).T
+    norm_cf = np.nan_to_num(x=norm_cf, nan=0.0)
+    sns.heatmap(norm_cf, annot=True, fmt="0.2f", xticklabels=_target.categories, ax=ax[1])
+    ax[1].set_xlabel("Target", size=labelsize)
+    ax[1].set_yticklabels(_preds.categories, rotation=270)
+    ax[1].set_ylabel("Prediction", size=labelsize)
 
-    _preds = pd.cut(x=preds, bins=bins, labels=labels)
-    _target = pd.cut(x=target, bins=bins, labels=labels)
-    breakpoint()
+    fig.suptitle(run_name, size=titlesize)
+    fig.tight_layout()
+    fig.savefig(root_dir / f"{run_name}-cf.png")
 
 
-def _cls_infrence(preds, target, root_dir, run_name) -> None:
+def _cls_inference(preds, target, root_dir, run_name) -> None:
     metrics_input = dict(preds=preds,
                          target=target.int(),
                          task="binary")