scitex 2.0.0__py2.py3-none-any.whl

scitex/ai/classification/classification_reporter.py

@@ -0,0 +1,1137 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# Timestamp: "2025-02-15 01:38:28 (ywatanabe)"
+# File: ./src/scitex/ai/ClassificationReporter.py
+
+THIS_FILE = "/home/ywatanabe/proj/scitex_repo/src/scitex/ai/ClassificationReporter.py"
+
+import os as _os
+import random as _random
+import sys as _sys
+from collections import defaultdict as _defaultdict
+from glob import glob as _glob
+from pprint import pprint as _pprint
+
+import matplotlib as _matplotlib
+import matplotlib.pyplot as _plt
+import scitex as _scitex
+import numpy as _np
+import pandas as _pd
+import torch as _torch
+from sklearn.metrics import (
+    balanced_accuracy_score as _balanced_accuracy_score,
+    classification_report as _classification_report,
+    confusion_matrix as _confusion_matrix,
+    matthews_corrcoef as _matthews_corrcoef,
+)
+
+# Lazy import to avoid circular dependency
+def _get_fix_seeds():
+    import scitex
+    return scitex.reproduce.fix_seeds
+
+_fix_seeds = None  # Will be loaded when needed
+
+
+class MultiClassificationReporter(object):
+    def __init__(self, sdir, tgts=None):
+        if tgts is None:
+            sdirs = [""]
+        else:
+            sdirs = [_os.path.join(sdir, tgt, "/") for tgt in tgts]
+        sdirs = [sdir + tgt + "/" for tgt in tgts]
+
+        self.tgt2id = {tgt: i_tgt for i_tgt, tgt in enumerate(tgts)}
+        self.reporters = [ClassificationReporter(sdir) for sdir in sdirs]
+
+    def add(self, obj_name, obj, tgt=None):
+        i_tgt = self.tgt2id[tgt]
+        self.reporters[i_tgt].add(obj_name, obj)
+
+    def calc_metrics(
+        self,
+        true_class,
+        pred_class,
+        pred_proba,
+        labels=None,
+        i_fold=None,
+        show=True,
+        auc_plt_config=dict(
+            figsize=(7, 7),
+            labelsize=8,
+            fontsize=7,
+            legendfontsize=6,
+            tick_size=0.8,
+            tick_width=0.2,
+        ),
+        tgt=None,
+    ):
+        i_tgt = self.tgt2id[tgt]
+        self.reporters[i_tgt].calc_metrics(
+            true_class,
+            pred_class,
+            pred_proba,
+            labels=labels,
+            i_fold=i_fold,
+            show=show,
+            auc_plt_config=auc_plt_config,
+        )
+
+    def summarize(
+        self,
+        n_round=3,
+        show=False,
+        tgt=None,
+    ):
+        i_tgt = self.tgt2id[tgt]
+        self.reporters[i_tgt].summarize(
+            n_round=n_round,
+            show=show,
+        )
+
+    def save(
+        self,
+        files_to_reproduce=None,
+        meta_dict=None,
+        tgt=None,
+    ):
+        i_tgt = self.tgt2id[tgt]
+        self.reporters[i_tgt].save(
+            files_to_reproduce=files_to_reproduce,
+            meta_dict=meta_dict,
+        )
+
+    def plot_and_save_conf_mats(
+        self,
+        plt,
+        extend_ratio=1.0,
+        colorbar=True,
+        confmat_plt_config=None,
+        sci_notation_kwargs=None,
+        tgt=None,
+    ):
+        i_tgt = self.tgt2id[tgt]
+        self.reporters[i_tgt].plot_and_save_conf_mats(
+            plt,
+            extend_ratio=extend_ratio,
+            colorbar=colorbar,
+            confmat_plt_config=confmat_plt_config,
+            sci_notation_kwargs=sci_notation_kwargs,
+        )
+
+
+class ClassificationReporter(object):
+    """Saves the following metrics under sdir.
+       - Balanced Accuracy
+       - MCC
+       - Confusion Matrix
+       - Classification Report
+       - ROC AUC score / curve
+       - PRE-REC AUC score / curve
+
+    Example is described in this file.
+    """
+
+    def __init__(self, sdir):
+        self.sdir = sdir
+        self.folds_dict = _defaultdict(list)
+        # Use lazy-loaded fix_seeds
+        fix_seeds_func = _fix_seeds or _get_fix_seeds()
+        fix_seeds_func(os=_os, random=_random, np=_np, torch=_torch, verbose=False)
+
+    def add(
+        self,
+        obj_name,
+        obj,
+    ):
+        """
+        ## fig
+        fig, ax = plt.subplots()
+        ax.plot(np.random.rand(10))
+        reporter.add("manu_figs", fig)
+
+        ## DataFrame
+        df = pd.DataFrame(np.random.rand(5, 3))
+        reporter.add("manu_dfs", df)
+
+        ## scalar
+        scalar = random.random()
+        reporter.add("manu_scalers", scalar)
+        """
+        assert isinstance(obj_name, str)
+        self.folds_dict[obj_name].append(obj)
+
+    @staticmethod
+    def calc_bACC(true_class, pred_class, i_fold, show=False):
+        """Balanced ACC"""
+        balanced_acc = _balanced_accuracy_score(true_class, pred_class)
+        if show:
+            print(f"\nBalanced ACC in fold#{i_fold} was {balanced_acc:.3f}\n")
+        return balanced_acc
+
+    @staticmethod
+    def calc_mcc(true_class, pred_class, i_fold, show=False):
+        """MCC"""
+        mcc = float(_matthews_corrcoef(true_class, pred_class))
+        if show:
+            print(f"\nMCC in fold#{i_fold} was {mcc:.3f}\n")
+        return mcc
+
+    @staticmethod
+    def calc_conf_mat(true_class, pred_class, labels, i_fold, show=False):
+        """
+        Confusion Matrix
+        This method assumes unique classes of true_class and pred_class are the same.
+        """
+        conf_mat = _pd.DataFrame(
+            data=_confusion_matrix(
+                true_class, pred_class, labels=_np.arange(len(labels))
+            ),
+            columns=labels,
+        ).set_index(_pd.Series(list(labels)))
+
+        if show:
+            print(f"\nConfusion Matrix in fold#{i_fold}: \n")
+            _pprint(conf_mat)
+            print()
+
+        return conf_mat
+
+    @staticmethod
+    def calc_clf_report(
+        true_class, pred_class, labels, balanced_acc, i_fold, show=False
+    ):
+        """Classification Report"""
+        clf_report = _pd.DataFrame(
+            _classification_report(
+                true_class,
+                pred_class,
+                labels=_np.arange(len(labels)),
+                target_names=labels,
+                output_dict=True,
+            )
+        )
+
+        clf_report["accuracy"] = balanced_acc
+        clf_report = _pd.concat(
+            [
+                clf_report[labels],
+                clf_report[["accuracy", "macro avg", "weighted avg"]],
+            ],
+            axis=1,
+        )
+        clf_report = clf_report.rename(columns={"accuracy": "balanced accuracy"})
+        clf_report = clf_report.round(3)
+        clf_report["index"] = clf_report.index
+        clf_report.loc["support", "index"] = "sample size"
+        clf_report.set_index("index", drop=True, inplace=True)
+        clf_report.index.name = None
+        if show:
+            print(f"\nClassification Report for fold#{i_fold}:\n")
+            _pprint(clf_report)
+            print()
+        return clf_report
+
+    def calc_AUCs(
+        self,
+        true_class,
+        pred_proba,
+        labels,
+        i_fold,
+        show=True,
+        auc_plt_config=dict(
+            figsize=(7, 7),
+            labelsize=8,
+            fontsize=7,
+            legendfontsize=6,
+            tick_size=0.8,
+            tick_width=0.2,
+        ),
+    ):
+        """ROC AUC and PRE-REC AUC."""
+        n_classes = len(labels)
+        assert len(_np.unique(true_class)) == n_classes
+        if n_classes == 2:
+            roc_auc = self._calc_AUCs_binary(
+                true_class,
+                pred_proba,
+                i_fold,
+                show=show,
+                auc_plt_config=auc_plt_config,
+            )
+        else:
+            roc_auc = self._calc_AUCs_multiple(
+                true_class,
+                pred_proba,
+                labels,
+                i_fold,
+                show=show,
+                auc_plt_config=auc_plt_config,
+            )
+        return roc_auc
+
+    def _calc_AUCs_binary(
+        self,
+        true_class,
+        pred_proba,
+        i_fold,
+        show=False,
+        auc_plt_config=dict(
+            figsize=(7, 7),
+            labelsize=8,
+            fontsize=7,
+            legendfontsize=6,
+            tick_size=0.8,
+            tick_width=0.2,
+        ),
+    ):
+        """Calculates metrics for binary classification."""
+        from sklearn.metrics import (
+            PrecisionRecallDisplay,
+            RocCurveDisplay,
+            auc,
+            precision_recall_curve,
+            roc_curve,
+        )
+
+        unique_classes = sorted(list(_np.unique(true_class)))
+        n_classes = len(unique_classes)
+        assert n_classes == 2, "This method is only for binary classification"
+
+        # ROC curve
+        fpr, tpr, _ = roc_curve(true_class, pred_proba)
+        roc_auc = auc(fpr, tpr)
+
+        fig_size = auc_plt_config["figsize"]
+        fontsize = auc_plt_config["fontsize"]
+        labelsize = auc_plt_config["labelsize"]
+        legendfontsize = auc_plt_config["legendfontsize"]
+        tick_size = auc_plt_config["tick_size"]
+        tick_width = auc_plt_config["tick_width"]
+
+        fig_roc, ax_roc = _plt.subplots(figsize=fig_size)
+        RocCurveDisplay(
+            fpr=fpr,
+            tpr=tpr,
+            roc_auc=roc_auc,
+        ).plot(ax=ax_roc)
+        ax_roc.plot([0, 1], [0, 1], "k:")
+        ax_roc.set_xlabel("False Positive Rate", fontsize=labelsize)
+        ax_roc.set_ylabel("True Positive Rate", fontsize=labelsize)
+        ax_roc.set_title("ROC Curve", fontsize=fontsize)
+        ax_roc.legend(fontsize=legendfontsize)
+        ax_roc.tick_params(
+            axis="both",
+            which="major",
+            labelsize=tick_size,
+            width=tick_width,
+        )
+        self.folds_dict["ROC_fig"].append(fig_roc)
+        if show:
+            print(f"\nROC AUC in fold#{i_fold} is {roc_auc:.3f}\n")
+
+        # PRE-REC curve
+        fig_prerec, ax_prerec = _plt.subplots(figsize=fig_size)
+        PrecisionRecallDisplay.from_predictions(
+            true_class,
+            pred_proba,
+            ax=ax_prerec,
+        )
+        ax_prerec.set_xlabel("Recall", fontsize=labelsize)
+        ax_prerec.set_ylabel("Precision", fontsize=labelsize)
+        ax_prerec.set_title("Precision-Recall Curve", fontsize=fontsize)
+        ax_prerec.legend(fontsize=legendfontsize)
+        ax_prerec.tick_params(
+            axis="both",
+            which="major",
+            labelsize=tick_size,
+            width=tick_width,
+        )
+        self.folds_dict["PRE_REC_fig"].append(fig_prerec)
+
+        return roc_auc
+
+
+# #!/usr/bin/env python3
+# # -*- coding: utf-8 -*-
+# # Time-stamp: "2024-11-20 00:15:08 (ywatanabe)"
+# # File: ./scitex_repo/src/scitex/ai/ClassificationReporter.py
+
+# THIS_FILE = "/home/ywatanabe/proj/scitex_repo/src/scitex/ai/ClassificationReporter.py"
+
+# #!/usr/bin/env python3
+# # -*- coding: utf-8 -*-
+# # Time-stamp: "2024-11-13 12:54:17 (ywatanabe)"
+# # File: ./scitex_repo/src/scitex/ai/ClassificationReporter.py
+
+# import os
+# import random
+# import sys
+# from collections import defaultdict as _defaultdict
+# from glob import glob as _glob
+# from pprint import pprint as _pprint
+
+# import matplotlib
+# import matplotlib.pyplot as plt
+# import scitex
+# import numpy as np
+# import pandas as pd
+# import torch
+# from sklearn.metrics import (
+#     balanced_accuracy_score,
+#     classification_report,
+#     confusion_matrix,
+#     matthews_corrcoef,
+# )
+
+# from ..reproduce import fix_seeds
+
+
+# class MultiClassificationReporter(object):
+#     def __init__(self, sdir, tgts=None):
+#         if tgts is None:
+#             sdirs = [""]
+#         else:
+#             sdirs = [os.path.join(sdir, tgt, "/") for tgt in tgts]
+#         sdirs = [sdir + tgt + "/" for tgt in tgts]
+
+#         self.tgt2id = {tgt: i_tgt for i_tgt, tgt in enumerate(tgts)}
+#         self.reporters = [ClassificationReporter(sdir) for sdir in sdirs]
+
+#     def add(self, obj_name, obj, tgt=None):
+#         i_tgt = self.tgt2id[tgt]
+#         self.reporters[i_tgt].add(obj_name, obj)
+
+#     def calc_metrics(
+#         self,
+#         true_class,
+#         pred_class,
+#         pred_proba,
+#         labels=None,
+#         i_fold=None,
+#         show=True,
+#         auc_plt_config=dict(
+#             figsize=(7, 7),
+#             labelsize=8,
+#             fontsize=7,
+#             legendfontsize=6,
+#             tick_size=0.8,
+#             tick_width=0.2,
+#         ),
+#         tgt=None,
+#     ):
+#         i_tgt = self.tgt2id[tgt]
+#         self.reporters[i_tgt].calc_metrics(
+#             true_class,
+#             pred_class,
+#             pred_proba,
+#             labels=labels,
+#             i_fold=i_fold,
+#             show=show,
+#             auc_plt_config=auc_plt_config,
+#         )
+
+#     def summarize(
+#         self,
+#         n_round=3,
+#         show=False,
+#         tgt=None,
+#     ):
+#         i_tgt = self.tgt2id[tgt]
+#         self.reporters[i_tgt].summarize(
+#             n_round=n_round,
+#             show=show,
+#         )
+
+#     def save(
+#         self,
+#         files_to_reproduce=None,
+#         meta_dict=None,
+#         tgt=None,
+#     ):
+#         i_tgt = self.tgt2id[tgt]
+#         self.reporters[i_tgt].save(
+#             files_to_reproduce=files_to_reproduce,
+#             meta_dict=meta_dict,
+#         )
+
+#     def plot_and_save_conf_mats(
+#         self,
+#         plt,
+#         extend_ratio=1.0,
+#         colorbar=True,
+#         confmat_plt_config=None,
+#         sci_notation_kwargs=None,
+#         tgt=None,
+#     ):
+#         i_tgt = self.tgt2id[tgt]
+#         self.reporters[i_tgt].plot_and_save_conf_mats(
+#             plt,
+#             extend_ratio=extend_ratio,
+#             colorbar=colorbar,
+#             confmat_plt_config=confmat_plt_config,
+#             sci_notation_kwargs=sci_notation_kwargs,
+#         )
+
+
+# class ClassificationReporter(object):
+#     """Saves the following metrics under sdir.
+#        - Balanced Accuracy
+#        - MCC
+#        - Confusion Matrix
+#        - Classification Report
+#        - ROC AUC score / curve
+#        - PRE-REC AUC score / curve
+
+#     Example is described in this file.
+#     """
+
+#     def __init__(self, sdir):
+#         self.sdir = sdir
+#         self.folds_dict = _defaultdict(list)
+#         fix_seeds(os=os, random=random, np=np, torch=torch, show=False)
+
+#     def add(
+#         self,
+#         obj_name,
+#         obj,
+#     ):
+#         """
+#         ## fig
+#         fig, ax = plt.subplots()
+#         ax.plot(np.random.rand(10))
+#         reporter.add("manu_figs", fig)
+
+#         ## DataFrame
+#         df = pd.DataFrame(np.random.rand(5, 3))
+#         reporter.add("manu_dfs", df)
+
+#         ## scalar
+#         scalar = random.random()
+#         reporter.add("manu_scalers", scalar)
+#         """
+#         assert isinstance(obj_name, str)
+#         self.folds_dict[obj_name].append(obj)
+
+#     @staticmethod
+#     def calc_bACC(true_class, pred_class, i_fold, show=False):
+#         """Balanced ACC"""
+#         balanced_acc = balanced_accuracy_score(true_class, pred_class)
+#         if show:
+#             print(f"\nBalanced ACC in fold#{i_fold} was {balanced_acc:.3f}\n")
+#         return balanced_acc
+
+#     @staticmethod
+#     def calc_mcc(true_class, pred_class, i_fold, show=False):
+#         """MCC"""
+#         mcc = float(matthews_corrcoef(true_class, pred_class))
+#         if show:
+#             print(f"\nMCC in fold#{i_fold} was {mcc:.3f}\n")
+#         return mcc
+
+#     @staticmethod
+#     def calc_conf_mat(true_class, pred_class, labels, i_fold, show=False):
+#         """
+#         Confusion Matrix
+#         This method assumes unique classes of true_class and pred_class are the same.
+#         """
+#         # conf_mat = pd.DataFrame(
+#         #     data=confusion_matrix(true_class, pred_class),
+#         #     columns=pred_labels,
+#         #     index=true_labels,
+#         # )
+
+#         conf_mat = pd.DataFrame(
+#             data=confusion_matrix(
+#                 true_class, pred_class, labels=np.arange(len(labels))
+#             ),
+#             columns=labels,
+#         ).set_index(pd.Series(list(labels)))
+
+#         if show:
+#             print(f"\nConfusion Matrix in fold#{i_fold}: \n")
+#             _pprint(conf_mat)
+#             print()
+
+#         return conf_mat
+
+#     @staticmethod
+#     def calc_clf_report(
+#         true_class, pred_class, labels, balanced_acc, i_fold, show=False
+#     ):
+#         """Classification Report"""
+#         clf_report = pd.DataFrame(
+#             classification_report(
+#                 true_class,
+#                 pred_class,
+#                 labels=np.arange(len(labels)),
+#                 target_names=labels,
+#                 output_dict=True,
+#             )
+#         )
+
+#         # ACC to bACC
+#         clf_report["accuracy"] = balanced_acc
+#         clf_report = pd.concat(
+#             [
+#                 clf_report[labels],
+#                 clf_report[["accuracy", "macro avg", "weighted avg"]],
+#             ],
+#             axis=1,
+#         )
+#         clf_report = clf_report.rename(
+#             columns={"accuracy": "balanced accuracy"}
+#         )
+#         clf_report = clf_report.round(3)
+#         # Renames 'support' to 'sample size'
+#         clf_report["index"] = clf_report.index
+#         clf_report.loc["support", "index"] = "sample size"
+#         clf_report.set_index("index", drop=True, inplace=True)
+#         clf_report.index.name = None
+#         if show:
+#             print(f"\nClassification Report for fold#{i_fold}:\n")
+#             _pprint(clf_report)
+#             print()
+#         return clf_report
+
+#     @staticmethod
+#     def calc_and_plot_roc_curve(
+#         true_class, pred_proba, labels, sdir_for_csv=None
+#     ):
+#         # ROC-AUC
+#         fig_roc, metrics_roc_auc_dict = scitex.ml.plt.roc_auc(
+#             plt,
+#             true_class,
+#             pred_proba,
+#             labels,
+#             sdir_for_csv=sdir_for_csv,
+#         )
+#         plt.close()
+#         return fig_roc, metrics_roc_auc_dict
+
+#     @staticmethod
+#     def calc_and_plot_pre_rec_curve(true_class, pred_proba, labels):
+#         # PRE-REC AUC
+#         fig_pre_rec, metrics_pre_rec_auc_dict = scitex.ml.plt.pre_rec_auc(
+#             plt, true_class, pred_proba, labels
+#         )
+#         plt.close()
+#         return fig_pre_rec, metrics_pre_rec_auc_dict
+
+#     def calc_metrics(
+#         self,
+#         true_class,
+#         pred_class,
+#         pred_proba,
+#         labels=None,
+#         i_fold=None,
+#         show=True,
+#         auc_plt_config=dict(
+#             figsize=(7, 7),
+#             labelsize=8,
+#             fontsize=7,
+#             legendfontsize=6,
+#             tick_size=0.8,
+#             tick_width=0.2,
+#         ),
+#     ):
+#         """
+#         Calculates ACC, Confusion Matrix, Classification Report, and ROC-AUC score on a fold.
+#         Metrics and curves will be kept in self.folds_dict.
+#         """
+
+#         ## Preparation
+#         # for convenience
+#         true_class = scitex.gen.torch_to_arr(true_class).astype(int).reshape(-1)
+#         pred_class = (
+#             scitex.gen.torch_to_arr(pred_class).astype(np.float64).reshape(-1)
+#         )
+#         pred_proba = scitex.gen.torch_to_arr(pred_proba).astype(np.float64)
+
+#         # for curves
+#         scitex.plt.configure_mpl(
+#             plt,
+#             **auc_plt_config,
+#         )
+
+#         ## Calc metrics
+#         # Balanced ACC
+#         bacc = self.calc_bACC(true_class, pred_class, i_fold, show=show)
+#         self.folds_dict["balanced_acc"].append(bacc)
+
+#         # MCC
+#         self.folds_dict["mcc"].append(
+#             self.calc_mcc(true_class, pred_class, i_fold, show=show)
+#         )
+
+#         # Confusion Matrix
+#         self.folds_dict["conf_mat/conf_mat"].append(
+#             self.calc_conf_mat(
+#                 true_class,
+#                 pred_class,
+#                 labels,
+#                 i_fold,
+#                 show=show,
+#             )
+#         )
+
+#         # Classification Report
+#         self.folds_dict["clf_report"].append(
+#             self.calc_clf_report(
+#                 true_class, pred_class, labels, bacc, i_fold, show=show
+#             )
+#         )
+
+#         ## Curves
+#         # ROC curve
+#         self.sdir_for_roc_csv = f"{self.sdir}roc/csv/"
+#         fig_roc, metrics_roc_auc_dict = self.calc_and_plot_roc_curve(
+#             true_class,
+#             pred_proba,
+#             labels,
+#             sdir_for_csv=self.sdir_for_roc_csv + f"fold#{i_fold}/",
+#         )
+#         self.folds_dict["roc/micro"].append(
+#             metrics_roc_auc_dict["roc_auc"]["micro"]
+#         )
+#         self.folds_dict["roc/macro"].append(
+#             metrics_roc_auc_dict["roc_auc"]["macro"]
+#         )
+#         self.folds_dict["roc/figs"].append(fig_roc)
+
+#         # PRE-REC curve
+#         fig_pre_rec, metrics_pre_rec_auc_dict = (
+#             self.calc_and_plot_pre_rec_curve(true_class, pred_proba, labels)
+#         )
+#         self.folds_dict["pre_rec/micro"].append(
+#             metrics_pre_rec_auc_dict["pre_rec_auc"]["micro"]
+#         )
+#         self.folds_dict["pre_rec/macro"].append(
+#             metrics_pre_rec_auc_dict["pre_rec_auc"]["macro"]
+#         )
+#         self.folds_dict["pre_rec/figs"].append(fig_pre_rec)
+
+#     @staticmethod
+#     def _mk_cv_index(n_folds):
+#         return [
+#             f"{n_folds}-folds_CV_mean",
+#             f"{n_folds}-fold_CV_std",
+#         ] + [f"fold#{i_fold}" for i_fold in range(n_folds)]
+
+#     def summarize_roc(
+#         self,
+#     ):
+
+#         folds_dirs = _glob(self.sdir_for_roc_csv + "fold#*")
+#         n_folds = len(folds_dirs)
+
+#         # get class names
+#         _csv_files = _glob(os.path.join(folds_dirs[0], "*"))
+#         classes_str = [
+#             csv_file.split("/")[-1].split(".csv")[0] for csv_file in _csv_files
+#         ]
+
+#         # dfs_classes = []
+#         # take mean and std by each class
+#         for cls_str in classes_str:
+
+#             fpaths_cls = [
+#                 os.path.join(fold_dir, f"{cls_str}.csv")
+#                 for fold_dir in folds_dirs
+#             ]
+
+#             ys = []
+#             roc_aucs = []
+#             for fpath_cls in fpaths_cls:
+#                 loaded_df = scitex.io.load(fpath_cls)
+#                 ys.append(loaded_df["y"])
+#                 roc_aucs.append(loaded_df["roc_auc"])
+#             ys = pd.concat(ys, axis=1)
+#             roc_aucs = pd.concat(roc_aucs, axis=1)
+
+#             df_cls = loaded_df[["x"]].copy()
+#             df_cls["y_mean"] = ys.mean(axis=1)
+#             df_cls["y_std"] = ys.std(axis=1)
+#             df_cls["roc_auc_mean"] = roc_aucs.mean(axis=1)
+#             df_cls["roc_auc_std"] = roc_aucs.std(axis=1)
+
+#             spath_cls = os.path.join(
+#                 self.sdir_for_roc_csv, f"k-fold_mean_std/{cls_str}.csv"
+#             )
+#             scitex.io.save(df_cls, spath_cls)
+#             # dfs_classes.append(df_cls)
+
+#     def summarize(
+#         self,
+#         n_round=3,
+#         show=False,
+#     ):
+#         """
+#         1) Take mean and std of scalars/pd.Dataframes for folds.
+#         2) Replace self.folds_dict with the summarized DataFrames.
+#         """
+#         self.summarize_roc()
+
+#         _n_folds_all = [
+#             len(self.folds_dict[k]) for k in self.folds_dict.keys()
+#         ]  # sometimes includes 0 because AUC curves are not always defined.
+#         self.n_folds_intended = max(_n_folds_all)
+
+#         for i_k, k in enumerate(self.folds_dict.keys()):
+#             n_folds = _n_folds_all[i_k]
+
+#             if n_folds != 0:
+#                 ## listed scalars
+#                 if is_listed_X(self.folds_dict[k], [float, int]):
+#                     mm = np.mean(self.folds_dict[k])
+#                     ss = np.std(self.folds_dict[k], ddof=1)
+#                     sr = pd.DataFrame(
+#                         data=[mm, ss] + self.folds_dict[k],
+#                         index=self._mk_cv_index(n_folds),
+#                         columns=[k],
+#                     )
+#                     self.folds_dict[k] = sr.round(n_round)
+
+#                 ## listed pd.DataFrames
+#                 elif is_listed_X(self.folds_dict[k], pd.DataFrame):
+#                     zero_df_for_mm = 0 * self.folds_dict[k][0].copy()
+#                     zero_df_for_ss = 0 * self.folds_dict[k][0].copy()
+
+#                     mm = (
+#                         zero_df_for_mm
+#                         + np.stack(self.folds_dict[k]).mean(axis=0)
+#                     ).round(n_round)
+
+#                     ss = (
+#                         zero_df_for_ss
+#                         + np.stack(self.folds_dict[k]).std(axis=0, ddof=1)
+#                     ).round(n_round)
+
+#                     self.folds_dict[k] = [mm, ss] + [
+#                         df_fold.round(n_round)
+#                         for df_fold in self.folds_dict[k]
+#                     ]
+
+#                     if show:
+#                         print(
+#                             "\n----------------------------------------\n"
+#                             f"\n{k}\n"
+#                             f"\n{n_folds}-fold-CV mean:\n"
+#                         )
+#                         _pprint(self.folds_dict[k][0])
+#                         print(f"\n\n{n_folds}-fold-CV std.:\n")
+#                         _pprint(self.folds_dict[k][1])
+#                         print("\n\n----------------------------------------\n")
+
+#                 ## listed figures
+#                 elif is_listed_X(self.folds_dict[k], matplotlib.figure.Figure):
+#                     pass
+
+#                 else:
+#                     print(f"{k} was not summarized")
+#                     print(type(self.folds_dict[k][0]))
+
+#     def save(
+#         self,
+#         files_to_reproduce=None,
+#         meta_dict=None,
+#     ):
+#         """
+#         1) Saves the content of self.folds_dict.
+#         2) Plots the colormap of confusion matrices and saves them.
+#         3) Saves passed meta_dict under self.sdir
+
+#         Example:
+#             meta_df_1 = pd.DataFrame(data=np.random.rand(3,3))
+#             meta_dict_1 = {"a": 0}
+#             meta_dict_2 = {"b": 0}
+#             meta_dict = {"meta_1.csv": meta_df_1,
+#                          "meta_1.yaml": meta_dict_1,
+#                          "meta_2.yaml": meta_dict_1,
+#             }
+
+#         """
+#         if meta_dict is not None:
+#             for k, v in meta_dict.items():
+#                 scitex.io.save(v, self.sdir + k)
+
+#         for k in self.folds_dict.keys():
+
+#             ## pd.Series / pd.DataFrame
+#             if isinstance(self.folds_dict[k], pd.Series) or isinstance(
+#                 self.folds_dict[k], pd.DataFrame
+#             ):
+#                 scitex.io.save(self.folds_dict[k], self.sdir + f"{k}.csv")
+
+#             ## listed pd.DataFrame
+#             elif is_listed_X(self.folds_dict[k], pd.DataFrame):
+#                 scitex.io.save(
+#                     self.folds_dict[k],
+#                     self.sdir + f"{k}.csv",
+#                     # indi_suffix=self.cv_index,
+#                     indi_suffix=self._mk_cv_index(len(self.folds_dict[k])),
+#                 )
+
+#             ## listed figures
+#             elif is_listed_X(self.folds_dict[k], matplotlib.figure.Figure):
+#                 for i_fold, fig in enumerate(self.folds_dict[k]):
+#                     scitex.io.save(
+#                         self.folds_dict[k][i_fold],
+#                         self.sdir + f"{k}/fold#{i_fold}.png",
+#                     )
+
+#             else:
+#                 print(f"{k} was not saved")
+#                 print(type(self.folds_dict[k]))
+
+#         if files_to_reproduce is not None:
+#             if isinstance(files_to_reproduce, list):
+#                 files_to_reproduce = [files_to_reproduce]
+#             for f in files_to_reproduce:
+#                 scitex.io.save(f, self.sdir)
+
+#     def plot_and_save_conf_mats(
+#         self,
+#         plt,
+#         extend_ratio=1.0,
+#         colorbar=True,
+#         confmat_plt_config=None,
+#         sci_notation_kwargs=None,
+#     ):
+#         def _inner_plot_conf_mat(
+#             plt,
+#             cm_df,
+#             title,
+#             extend_ratio=1.0,
+#             colorbar=True,
+#             sci_notation_kwargs=None,
+#         ):
+#             labels = list(cm_df.columns)
+#             fig_conf_mat = scitex.ml.plt.confusion_matrix(
+#                 plt,
+#                 cm_df.T,
+#                 labels=labels,
+#                 title=title,
+#                 x_extend_ratio=extend_ratio,
+#                 y_extend_ratio=extend_ratio,
+#                 colorbar=colorbar,
+#             )
+
+#             if sci_notation_kwargs is not None:
+#                 fig_conf_mat.axes[-1] = scitex.plt.ax_scientific_notation(
+#                     fig_conf_mat.axes[-1], **sci_notation_kwargs
+#                 )
+#             return fig_conf_mat
+
+#         ## Configures mpl
+#         scitex.plt.configure_mpl(
+#             plt,
+#             **confmat_plt_config,
+#         )
+
+#         ########################################
+#         ## Prepares confmats dfs
+#         ########################################
+#         ## Drops mean and std for the folds
+#         try:
+#             conf_mats = self.folds_dict["conf_mat/conf_mat"][
+#                 -self.n_folds_intended :
+#             ]
+
+#         except Exception as e:
+#             print(e)
+#             conf_mats = self.folds_dict["conf_mat/conf_mat"]
+
+#         ## Prepaires conf_mat_overall_sum
+#         conf_mat_zero = 0 * conf_mats[0].copy()  # get the table format
+#         conf_mat_overall_sum = conf_mat_zero + np.stack(conf_mats).sum(axis=0)
+
+#         ########################################
+#         ## Plots & Saves
+#         ########################################
+#         # each fold's conf
+#         for i_fold, cm in enumerate(conf_mats):
+#             title = f"Test fold#{i_fold}"
+#             fig_conf_mat_fold = _inner_plot_conf_mat(
+#                 plt,
+#                 cm,
+#                 title,
+#                 extend_ratio=extend_ratio,
+#                 colorbar=colorbar,
+#                 sci_notation_kwargs=sci_notation_kwargs,
+#             )
+#             scitex.io.save(
+#                 fig_conf_mat_fold,
+#                 self.sdir + f"conf_mat/figs/fold#{i_fold}.png",
+#             )
+#             plt.close()
+
+#         ## overall_sum conf_mat
+#         title = f"{self.n_folds_intended}-CV overall sum"
+#         fig_conf_mat_overall_sum = _inner_plot_conf_mat(
+#             plt,
+#             conf_mat_overall_sum,
+#             title,
+#             extend_ratio=extend_ratio,
+#             colorbar=colorbar,
+#             sci_notation_kwargs=sci_notation_kwargs,
+#         )
+#         scitex.io.save(
+#             fig_conf_mat_overall_sum,
+#             self.sdir
+#             + f"conf_mat/figs/{self.n_folds_intended}-fold_cv_overall-sum.png",
+#         )
+#         plt.close()
+
+
+# if __name__ == "__main__":
+#     import random
+#     import sys
+
+#     import scitex
+#     import numpy as np
+#     from catboost import CatBoostClassifier, Pool
+#     from sklearn.datasets import load_digits
+#     from sklearn.model_selection import StratifiedKFold
+
+#     ################################################################################
+#     ## Sets tee
+#     ################################################################################
+#     sdir = scitex.io.mk_spath(
+#         "./tmp/sdir-ClassificationReporter/"
+#     )  # "/tmp/sdir/"
+#     sys.stdout, sys.stderr = scitex.gen.tee(sys, sdir)
+
+#     ################################################################################
+#     ## Fixes seeds
+#     ################################################################################
+#     fix_seeds(np=np)
+
+#     ## Loads
+#     mnist = load_digits()
+#     X, T = mnist.data, mnist.target
+#     labels = mnist.target_names.astype(str)
+
+#     ## Main
+#     skf = StratifiedKFold(n_splits=5, shuffle=True)
+#     # reporter = ClassificationReporter(sdir)
+#     mreporter = MultiClassificationReporter(sdir, tgts=["Test1", "Test2"])
+#     for i_fold, (indi_tra, indi_tes) in enumerate(skf.split(X, T)):
+#         X_tra, T_tra = X[indi_tra], T[indi_tra]
+#         X_tes, T_tes = X[indi_tes], T[indi_tes]
+
+#         clf = CatBoostClassifier(verbose=False)
+
+#         clf.fit(X_tra, T_tra, verbose=False)
+
+#         ## Prediction
+#         pred_proba_tes = clf.predict_proba(X_tes)
+#         pred_cls_tes = np.argmax(pred_proba_tes, axis=1)
+
+#         pred_cls_tes[pred_cls_tes == 9] = 8  # overide 9 as 8 # fixme
+
+#         ##############################
+#         ## Manually adds objects to reporter to save
+#         ##############################
+#         ## Figure
+#         fig, ax = plt.subplots()
+#         ax.plot(np.arange(10))
+#         # reporter.add("manu_figs", fig)
+#         mreporter.add("manu_figs", fig, tgt="Test1")
+#         mreporter.add("manu_figs", fig, tgt="Test2")
+
+#         ## DataFrame
+#         df = pd.DataFrame(np.random.rand(5, 3))
+#         # reporter.add("manu_dfs", df)
+#         mreporter.add("manu_dfs", df, tgt="Test1")
+#         mreporter.add("manu_dfs", df, tgt="Test2")
+
+#         ## Scalar
+#         scalar = random.random()
+#         # reporter.add(
+#         #     "manu_scalars",
+#         #     scalar,
+#         # )
+#         mreporter.add("manu_scalars", scalar, tgt="Test1")
+#         mreporter.add("manu_scalars", scalar, tgt="Test2")
+
+#         ########################################
+#         ## Metrics
+#         ########################################
+#         mreporter.calc_metrics(
+#             T_tes,
+#             pred_cls_tes,
+#             pred_proba_tes,
+#             labels=labels,
+#             i_fold=i_fold,
+#             tgt="Test1",
+#         )
+#         mreporter.calc_metrics(
+#             T_tes,
+#             pred_cls_tes,
+#             pred_proba_tes,
+#             labels=labels,
+#             i_fold=i_fold,
+#             tgt="Test2",
+#         )
+
+#     # reporter.summarize(show=True)
+#     mreporter.summarize(show=True, tgt="Test1")
+#     mreporter.summarize(show=True, tgt="Test2")
+
+#     fake_fpaths = ["fake_file_1.txt", "fake_file_2.txt"]
+#     for ff in fake_fpaths:
+#         scitex.io.touch(ff)
+
+#     files_to_reproduce = [
+#         scitex.gen.get_this_fpath(when_ipython="/dev/null"),
+#         *fake_fpaths,
+#     ]
+#     # reporter.save(files_to_reproduce=files_to_reproduce)
+#     mreporter.save(files_to_reproduce=files_to_reproduce, tgt="Test1")
+#     mreporter.save(files_to_reproduce=files_to_reproduce, tgt="Test2")
+
+#     confmat_plt_config = dict(
+#         figsize=(8, 8),
+#         # labelsize=8,
+#         # fontsize=6,
+#         # legendfontsize=6,
+#         figscale=2,
+#         tick_size=0.8,
+#         tick_width=0.2,
+#     )
+
+#     sci_notation_kwargs = dict(
+#         order=1,
+#         fformat="%1.0d",
+#         scilimits=(-3, 3),
+#         x=False,
+#         y=True,
+#     )  # "%3.1f"
+
+#     # sci_notation_kwargs = None
+#     # reporter.plot_and_save_conf_mats(
+#     #     plt,
+#     #     extend_ratio=1.0,
+#     #     confmat_plt_config=confmat_plt_config,
+#     #     sci_notation_kwargs=sci_notation_kwargs,
+#     # )
+
+#     mreporter.plot_and_save_conf_mats(
+#         plt,
+#         extend_ratio=1.0,
+#         confmat_plt_config=confmat_plt_config,
+#         sci_notation_kwargs=sci_notation_kwargs,
+#         tgt="Test1",
+#     )
+#     mreporter.plot_and_save_conf_mats(
+#         plt,
+#         extend_ratio=1.0,
+#         confmat_plt_config=confmat_plt_config,
+#         sci_notation_kwargs=sci_notation_kwargs,
+#         tgt="Test2",
+#     )
+
+# python -m scitex.ai.ClassificationReporter
+
+# EOF