Mentions légales du service

Skip to content
Snippets Groups Projects
analysis.py 48.1 KiB
Newer Older
DEBREUVE Eric's avatar
DEBREUVE Eric committed
# Copyright CNRS/Inria/UNS
# Contributor(s): Eric Debreuve (since 2019), Morgane Nadal (2020)
#
# eric.debreuve@cnrs.fr
#
# This software is governed by the CeCILL  license under French law and
# abiding by the rules of distribution of free software.  You can  use,
# modify and/ or redistribute the software under the terms of the CeCILL
# license as circulated by CEA, CNRS and INRIA at the following URL
# "http://www.cecill.info".
#
# As a counterpart to the access to the source code and  rights to copy,
# modify and redistribute granted by the license, users are provided only
# with a limited warranty  and the software's author,  the holder of the
# economic rights,  and the successive licensors  have only  limited
# liability.
#
# In this respect, the user's attention is drawn to the risks associated
# with loading,  using,  modifying and/or developing or reproducing the
# software by the user in light of its specific status of free software,
# that may mean  that it is complicated to manipulate,  and  that  also
# therefore means  that it is reserved for developers  and  experienced
# professionals having in-depth computer knowledge. Users are therefore
# encouraged to load and test the software's suitability as regards their
# requirements in conditions enabling the security of their systems and/or
# data to be ensured and,  more generally, to use and operate it in the
# same conditions as regards security.
#
# The fact that you are presently reading this means that you have had
# knowledge of the CeCILL license and that you accept its terms.

import itertools
import sys as sy_
from pathlib import Path as path_t
DEBREUVE Eric's avatar
DEBREUVE Eric committed
import matplotlib.pyplot as pl_
import numpy as nmpy
import pandas as pd_
import scipy as si_
import seaborn as sb_
from logger_36 import LOGGER
DEBREUVE Eric's avatar
DEBREUVE Eric committed
from matplotlib.lines import Line2D
from sklearn import tree
DEBREUVE Eric's avatar
DEBREUVE Eric committed
from sklearn.cluster import KMeans
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler


    colors: List[str],
    save_name: str = None,
    title: str = "",
    plot_duration: bool = True,
    save_in: str = None,
    three_D: bool = False,
):
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    Perform 2D or 3D PCA on the CHO-DIO dataframe.
    Save explained variance ratio as csv, plot and save the PCAs.
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    # Separating the features from their conditions and durations
    all_target = pd_.DataFrame(df.loc[:, [target]].values, columns=[target])
    df_all = df.drop([target, "duration"], axis=1)

    # Standardize the data
    scaler = StandardScaler()
    scaler.fit(df_all)
    stand_df = scaler.transform(df_all)

    # Create the PCA and fit the data
    pca = PCA(n_components=2)
    principal_components = pca.fit_transform(stand_df)

    # Printing and saving the explained variance ratio
    LOGGER.info(
        f"PCA explained variance ratio ({save_name}): {pca.explained_variance_ratio_}"
    )
    pca_exp_var_df = pd_.DataFrame(
        pca.explained_variance_ratio_, columns=["pca explained variance ration"]
    )
    pca_exp_var_df = pca_exp_var_df.rename(
        index={0: "principal component 1", 1: "principal component 2"}
    )
    pca_exp_var_df.to_csv(
        str(path_t(save_in) / f"pca_explained_variance_ratio_{save_name}.csv")
    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

    # Creating the principal component df
    principal_df = pd_.DataFrame(
        data=principal_components,
        columns=["principal component 1", "principal component 2"],
    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    # Give the final df containing the principal component and their condition
    final_df = pd_.concat([principal_df, all_target[target]], axis=1)

    # Plot 2D
    fig = pl_.figure(figsize=(8, 8))
    ax = fig.add_subplot(1, 1, 1)
    ax.set_xlabel(
        f"Principal Component 1 - ratio = {round(pca.explained_variance_ratio_[0],2)}",
        fontsize=15,
    )
    ax.set_ylabel(
        f"Principal Component 2 - ratio = {round(pca.explained_variance_ratio_[1],2)}",
        fontsize=15,
    )
    ax.set_title(f"2 component PCA{title}", fontsize=20)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    for tgt, color in zip(targets, colors):
        idx = final_df[target] == tgt
        ax.scatter(
            final_df.loc[idx, "principal component 1"],
            final_df.loc[idx, "principal component 2"],
            c=color,
            s=30,
        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    ax.legend(targets)
    ax.grid()

    if save_name is not None:
        pl_.savefig(str(path_t(save_in) / f"PCA_{save_name}.png"))
DEBREUVE Eric's avatar
DEBREUVE Eric committed

DEBREUVE Eric's avatar
DEBREUVE Eric committed
    if plot_duration:
        # Make separated plots for each duration of the experiments
        fig = pl_.figure(figsize=(8, 8))
        ax = fig.add_subplot(1, 1, 1)

        ax.set_xlabel(
            f"Principal Component 1 - ratio = {round(pca.explained_variance_ratio_[0],2)}",
            fontsize=15,
        )
        ax.set_ylabel(
            f"Principal Component 2 - ratio = {round(pca.explained_variance_ratio_[1],2)}",
            fontsize=15,
        )
        ax.set_title(f"2 component PCA{title}", fontsize=20)

        # new_tgts = [
        #     ["CHO", "1H"],
        #     ["CHO", "3H"],
        #     ["CHO", "6H"],
        #     ["DIO", "1H"],
        #     ["DIO", "3H"],
        #     ["DIO", "6H"],
        # ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        final_df = pd_.concat([df[["condition", "duration"]], principal_df], axis=1)
        new_tgts = _Experiments(df)
        LOGGER.info(f"New Targets: {new_tgts}")
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        for tgt, color in zip(
            new_tgts,
            ["lavender", "royalblue", "navy", "lightcoral", "firebrick", "red"],
        ):
            new_df = final_df.loc[
                (final_df["condition"] == tgt[0]) & (final_df["duration"] == tgt[1])
            ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            new_df = new_df.drop(["condition", "duration"], axis=1)

            ax.scatter(
                new_df["principal component 1"],
                new_df["principal component 2"],
                c=color,
                s=30,
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        ax.legend(new_tgts)
        ax.grid()

        if save_name is not None:
            pl_.savefig(str(path_t(save_in) / f"PCA_duration_{save_name}.png"))
DEBREUVE Eric's avatar
DEBREUVE Eric committed

    if three_D:
        # Create the 3D PCA and fit the data
        pca3d = PCA(n_components=3)
        principal_components3d = pca3d.fit_transform(stand_df)

        # Print and Save the explained variance ratio
            f"3D PCA explained variance ratio ({save_name}): {pca3d.explained_variance_ratio_}"
        )
        pca3d_exp_var_df = pd_.DataFrame(
            pca3d.explained_variance_ratio_,
            columns=["3d pca explained variance ration"],
        )
        pca3d_exp_var_df = pca3d_exp_var_df.rename(
            index={
                0: "principal component 1",
                1: "principal component 2",
                2: "principal component 3",
            }
        )
        pca3d_exp_var_df.to_csv(
            str(
                path_t(save_in)
                / f"three_d_pca_explained_variance_ratio_{save_name}.csv"
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        # Creating the principal component df
        principal3d_df = pd_.DataFrame(
            data=principal_components3d,
            columns=[
                "principal component 1",
                "principal component 2",
                "principal component 3",
            ],
        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Give the final df containing the principal component and their condition
        final3d_df = pd_.concat([principal3d_df, all_target[target]], axis=1)

        # Plot
        fig = pl_.figure(figsize=(8, 8))
        ax = fig.add_subplot(111, projection="3d")

        ax.set_xlabel(
            f"Principal Component 1 - ratio = {round(pca3d.explained_variance_ratio_[0],2)}",
            fontsize=15,
        )
        ax.set_ylabel(
            f"Principal Component 2 - ratio = {round(pca3d.explained_variance_ratio_[1],2)}",
            fontsize=15,
        )
        ax.set_zlabel(
            f"Principal Component 3 - ratio = {round(pca3d.explained_variance_ratio_[2],2)}",
            fontsize=15,
        )
        ax.set_title(f"3 component PCA{title}", fontsize=20)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        for tgt, color in zip(targets, colors):
            idx = final3d_df[target] == tgt

            ax.scatter(
                final3d_df.loc[idx, "principal component 1"],
                final3d_df.loc[idx, "principal component 2"],
                final3d_df.loc[idx, "principal component 3"],
                c=color,
                s=30,
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        ax.legend(targets)
        ax.grid()

        if save_name is not None:
            pl_.savefig(str(path_t(save_in) / f"three_d_PCA_{save_name}.png"))
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        if plot_duration:
            fig = pl_.figure(figsize=(8, 8))
            ax = fig.add_subplot(111, projection="3d")

            ax.set_xlabel(
                f"Principal Component 1 - ratio = {round(pca3d.explained_variance_ratio_[0], 2)}",
                fontsize=15,
            )
            ax.set_ylabel(
                f"Principal Component 2 - ratio = {round(pca3d.explained_variance_ratio_[1],2)}",
                fontsize=15,
            )
            ax.set_zlabel(
                f"Principal Component 3 - ratio = {round(pca3d.explained_variance_ratio_[2],2)}",
                fontsize=15,
            )
            ax.set_title(f"3 component PCA{title}", fontsize=20)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

            final3d_df = pd_.concat(
                [df[["condition", "duration"]], principal3d_df], axis=1
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

            for tgt, color in zip(
                new_tgts,
                ["lavender", "royalblue", "navy", "lightcoral", "firebrick", "red"],
            ):
                new3d_df = final3d_df.loc[
                    (final_df["condition"] == tgt[0])
                    & (final3d_df["duration"] == tgt[1])
                ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                new3d_df = new3d_df.drop(["condition", "duration"], axis=1)

                ax.scatter(
                    new3d_df["principal component 1"],
                    new3d_df["principal component 2"],
                    new3d_df["principal component 3"],
                    c=color,
                    s=30,
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

            ax.legend(new_tgts)
            ax.grid()

            if save_name is not None:
                pl_.savefig(
                    str(path_t(save_in) / f"three_d_PCA_duration_{save_name}.png")
                )
    nb_clusters: tuple,
    target: str,
    plot_bar: bool = True,
    rep_on_image: bool = False,
    labeled_somas=None,
    elbow: bool = False,
    intracluster_var: bool = True,
    save_name: str = None,
    save_in: str = None,
    title: str = "",
    duration: bool = False,
    features_distribution: bool = True,
) -> KMeans:
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    Perform kmeans on the pandas dataframe. Can find the best number of cluster with elbow method,
    find the intracluster variance, and represent the result on the initial images.
    Plot barplots with percentage of each label for each condition.
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    # Separating the features from their conditions and durations
    all_target = pd_.DataFrame(df.loc[:, [target]].values, columns=[target])
    df2 = df.drop([target, "duration"], axis=1)
    df_scalar = df.drop(["duration", "condition"], axis=1)

    # Data standardization
    scaler = StandardScaler()
    stand_df = scaler.fit_transform(df2)

DEBREUVE Eric's avatar
DEBREUVE Eric committed
    # Best number of clusters using Elbow method
    if elbow:
        wcss = []  # within cluster sum of errors(wcss)
        for i in range(1, 24):
            kmeans = KMeans(n_clusters=i, n_init=10, random_state=0)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            kmeans.fit(stand_df)
            wcss.append(kmeans.inertia_)
        pl_.figure()
        pl_.plot(range(1, 24), wcss)
        pl_.plot(range(1, 24), wcss, "bo")
        pl_.title("Elbow Method")
        pl_.xlabel("Number of clusters")
        pl_.ylabel("WCSS")
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        pl_.show(block=True)
        pl_.close()

    # Kmeans with x clusters
    for nb_cluster in nb_clusters:
        kmeans = KMeans(n_clusters=nb_cluster, n_init=10, random_state=0)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        kmeans.fit_predict(stand_df)
        label_df = pd_.DataFrame(data=kmeans.labels_, columns=["label"])
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        lab_cond_df = pd_.concat([label_df, all_target[target]], axis=1)

        # Intracluster variance
        if intracluster_var:
            var_df = pd_.DataFrame(stand_df)
            var = IntraClusterVariance(var_df, kmeans, nb_cluster, save_name)
            var_df = pd_.DataFrame(var, columns=["intracluster variance"])
            var_df = var_df.rename(
                {cluster: f"label {cluster}" for cluster in range(nb_cluster)}
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            if save_in:
                    str(
                        path_t(save_in)
                        / f"intracluster_variance_kmeans_{save_name}_k={nb_cluster}.csv"
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        # TODO Intersection of ellipsoids of the covariance matrices
        #  - ellipsoid equation: (x-mean)(1/Mcov)(x-mean).T <= 1

        # Barplot
        if plot_bar:
            if duration:
                fig = pl_.figure(figsize=(16, 8))
                ax = fig.add_subplot(1, 1, 1)
                (
                    lab_cond_df.groupby("condition")["label"]
                    .value_counts(normalize=True)
                    .mul(100)
                    .rename("Percentage (%)")
                    .reset_index()
                    .pipe(
                        (sb_.barplot, "data"),
                        x="condition",
                        y="Percentage (%)",
                        hue="label",
                        palette=sb_.color_palette("deep", n_colors=nb_cluster),
                    )
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                ax.set_ylim(0, 100)
                ax.set_title(
                    f"Distribution of the clustering labels according to conditions{title}_k={nb_cluster}",
                    fontsize=11,
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                ax.grid()
                if save_name is not None:
                    pl_.savefig(
                        str(
                            path_t(save_in)
                            / f"Hist_Clustering_{save_name}_k={nb_cluster}.png"
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    # pl_.show(block=True)
                    # pl_.close()
            else:
                ## Batthacharyya similarity
                groupbycond = lab_cond_df.groupby("condition")
                hist = []
                for cond, values in groupbycond:
                    hist.append(nmpy.histogram(values["label"], bins=2))
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                bhatt = BhattacharyyaSimilarity(hist[0][0], hist[1][0])
                LOGGER.info(
                    f"Bhattacharyya similarity btw CHO and DIO, all times: {bhatt}"
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                ## Plot the barplots
                fig = pl_.figure(figsize=(16, 8))
                ax = fig.add_subplot(1, 1, 1)
                (
                    groupbycond["label"]
                    .value_counts(normalize=True)
                    .mul(100)
                    .rename("Percentage (%)")
                    .reset_index()
                    .pipe(
                        (sb_.barplot, "data"),
                        x="condition",
                        y="Percentage (%)",
                        hue="label",
                        palette=sb_.color_palette("deep", n_colors=nb_cluster),
                    )
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                ax.set_ylim(0, 100)
                ax.grid()
                ax.set_title(
                    f"Distribution of the clustering labels according to conditions{title}_k={nb_cluster}",
                    fontsize=11,
                )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                if save_name is not None:
                    pl_.savefig(
                        str(
                            path_t(save_in)
                            / f"Hist_Clustering_{save_name}_k={nb_cluster}.png"
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    # pl_.show(block=True)
                    # pl_.close()
                lab_cond_df2 = pd_.concat((lab_cond_df, df[["duration"]]), axis=1)

                bhattacharyya = []
                for dur in df["duration"].unique():
                    dur_df = lab_cond_df2.loc[lab_cond_df2["duration"] == dur]

                    ## Bhattacharyya
                    groupbycond = dur_df.groupby("condition")
                    hist = []
                    for cond, values in groupbycond:
                        hist.append(nmpy.histogram(values["label"], bins=2))
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    bhatt = BhattacharyyaSimilarity(hist[0][0], hist[1][0])
                    LOGGER.info(
                        f"Bhattacharyya similarity btw CHO and DIO {dur}: {bhatt}"
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    bhattacharyya.append(bhatt)

                    ## Barplot
                    fig = pl_.figure(figsize=(16, 8))
                    ax = fig.add_subplot(1, 1, 1)
                    to_plot = (
                        dur_df.groupby("condition")["label"]
                        .value_counts(normalize=True)
                        .mul(100)
                        .rename("Percentage (%)")
                        .reset_index()
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    n_labels = to_plot["label"].unique()
                    n_labels.sort()
                    colors = sb_.color_palette("deep", n_colors=nb_cluster)
                    color_to_use = list(colors[i] for i in n_labels)
                    (
                        to_plot.pipe(
                            (sb_.barplot, "data"),
                            x="condition",
                            y="Percentage (%)",
                            hue="label",
                            palette=color_to_use,
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    ax.set_ylim(0, 100)
                    ax.grid()
                    ax.set_title(
                        f"Distribution of the clustering labels according to conditions{title}_k={nb_cluster} - duration = {dur}",
                        fontsize=11,
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    if save_name is not None:
                        pl_.savefig(
                            str(
                                path_t(save_in)
                                / f"Hist_Clustering_{save_name}_dur_{dur}_k={nb_cluster}.png"
                            )
                        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                        # pl_.show(block=True)
                        pl_.close()

                if bhattacharyya is not None:
                    fig = pl_.figure(figsize=(16, 8))
                    ax = fig.add_subplot(1, 1, 1)
                    ax.plot(
                        [f"{duration}" for duration in df["duration"].unique()],
                        [bhattacharyya[i] for i in range(3)],
                        "ko",
                    )
                    ax.plot(
                        [f"{duration}" for duration in df["duration"].unique()],
                        [bhattacharyya[i] for i in range(3)],
                        "g",
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    ax.set_xlabel("Duration")
                    ax.set_yscale("log")
                    ax.set_ylabel("Bhattacharyya distance")
                    ax.grid()
                    ax.set_title(
                        f"Bhattacharyya distance between CHO and DIO{title}_k={nb_cluster}",
                        fontsize=11,
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                    if save_name is not None:
                        pl_.savefig(
                            str(
                                path_t(save_in)
                                / f"Bhattacharyya_distance_{save_name}_k={nb_cluster}.png"
                            )
                        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                        pl_.close()

                if save_name is not None:
                    pl_.savefig(
                        str(
                            path_t(save_in)
                            / f"Hist_duration_Clustering_{save_name}_k={nb_cluster}.png"
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    print(f"Saved in {save_in}")
                    # pl_.show(block=True)
                    pl_.close()

        if features_distribution:
            for column in df_scalar.columns:
                cond_col_df = pd_.concat((df_scalar[column], lab_cond_df), axis=1)

                conditions = _Conditions(df_scalar)
                print("----", conditions)
                labels = nmpy.arange(nb_cluster)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                fig = pl_.figure(constrained_layout=False)
                gs = fig.add_gridspec(ncols=1, nrows=1)
                ax1 = fig.add_subplot(gs[0, 0])

                # Plot a histogram and kernel density estimate
                # print(f"Plot a histogram and kernel density estimate for feature {column}")
                for comb, color in zip(
                    itertools.product(conditions, labels),
                    sb_.color_palette("deep", n_colors=len(labels) * len(conditions)),
                ):
                    to_plot = cond_col_df.loc[(cond_col_df["condition"] == comb[0])]
                    to_plot = to_plot.loc[(to_plot["label"] == comb[1])]
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                    # Kernel estimate of the histogram
                    sb_.kdeplot(to_plot[[column]], color=color, ax=ax1)

                lines = [
                    Line2D([0], [0], color=c, linewidth=3, linestyle="-")
                    for c in sb_.color_palette(
                        "deep", n_colors=len(labels) * len(conditions)
                    )
                ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                lb = list(itertools.product(conditions, labels))
                ax1.legend(lines, lb)
                ax1.set_title(f"{column} distribution{title}", fontsize=11)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                pl_.tight_layout()

                if save_in is not None:
                    pl_.savefig(
                        str(
                            path_t(save_in)
                            / f"feat_kernel_estimate_{column}_k={nb_cluster}.png"
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    pl_.close()

                # Boxplot
                for comb, color in zip(
                    itertools.product(conditions, labels),
                    sb_.color_palette("deep", n_colors=len(labels) * len(conditions)),
                ):
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    fig = pl_.figure(constrained_layout=False, figsize=(10, 3))
                    gs = fig.add_gridspec(ncols=1, nrows=1)
                    ax1 = fig.add_subplot(gs[0, 0])

                    to_plot = cond_col_df.loc[(cond_col_df["condition"] == comb[0])]
                    to_plot = to_plot.loc[(to_plot["label"] == comb[1])]
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                    sb_.boxplot(to_plot[[column]], color=color, ax=ax1)

                    ax1.set_xlim(min(cond_col_df[column]), max(cond_col_df[column]))
                    lines = [Line2D([0], [0], color=color, linewidth=3, linestyle="-")]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    ax1.legend(lines, [comb])
                    ax1.set_title(f"{column} boxplot{title}", fontsize=11)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                    pl_.tight_layout()

                    if save_in is not None:
                        pl_.savefig(
                            str(
                                path_t(save_in)
                                / f"feat_boxplot_{column}_{comb}_k={nb_cluster}.png"
                            )
                        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                        pl_.close()

                # Do the same thing but separating durations
                cond_col_df_dur = pd_.concat((cond_col_df, df["duration"]), axis=1)
                groupby_dur_ = cond_col_df_dur.groupby("duration")

                fig = pl_.figure(constrained_layout=False, figsize=(15, 10))
                gs = fig.add_gridspec(ncols=3, nrows=1)
                ax1 = fig.add_subplot(gs[0, 0])
                ax2 = fig.add_subplot(gs[0, 1])
                ax3 = fig.add_subplot(gs[0, 2])
                ax = [ax1, ax2, ax3]

                # Plot a histogram and kernel density estimate
                # print(f"Plot a histogram and kernel density estimate for feature {column}")
                x = 0
                for dur_, val in groupby_dur_:
                    for comb, color in zip(
                        itertools.product(conditions, labels),
                        sb_.color_palette(
                            "deep", n_colors=len(labels) * len(conditions)
                        ),
                    ):
                        to_plot = val.loc[(val["condition"] == comb[0])]
                        to_plot = to_plot.loc[(to_plot["label"] == comb[1])]
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                        # Kernel estimate of the histogram
                        sb_.kdeplot(
                            to_plot[[column]], color=color, ax=ax[x]
                        )

                    lines = [
                        Line2D([0], [0], color=c, linewidth=3, linestyle="-")
                        for c in sb_.color_palette(
                            "deep", n_colors=len(labels) * len(conditions)
                        )
                    ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    lb = list(itertools.product(conditions, labels))
                    ax[x].legend(lines, lb)
                    ax[x].set_title(f"{dur_}", fontsize=11)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    # ax[x].set_xlim(min(val[column]), max(val[column]))
                    # ax[x].set_ylim()
                    ax[x].set_xlabel("Distribution kernel estimate")
                    ax[x].set_ylabel("Features values")
                    x += 1

                fig.suptitle(f"{column} distribution{title}")
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                # pl_.tight_layout()

                if save_in is not None:
                    pl_.savefig(
                        str(
                            path_t(save_in)
                            / f"feat_kernel_estimate_{column}_k={nb_cluster}_dur.png"
                        )
                    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                    pl_.close()

        # Representation on the image
        if rep_on_image:
            RepresentationOnImages(labeled_somas, kmeans, nb_cluster)

    return kmeans


def IntraClusterVariance(
    df: pd_.DataFrame, kmeans: KMeans(), nb_cluster: int, save_name: str = ""
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    Return the intracluster variance of a given cluster found by kmeans.
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    var = []
    # TODO change to .inertia_
    for cluster in range(nb_cluster):
        soma_cluster = [
            indx for indx, value in enumerate(kmeans.labels_) if value == cluster
        ]
        mean_cluster = nmpy.average(
            [df.iloc[row, :] for row in soma_cluster], axis=0
        )  # TODO .cluster_centers_
        variance = sum(
            [
                nmpy.linalg.norm(df.iloc[row, :] - mean_cluster) ** 2
                for row in soma_cluster
            ]
        ) / (len(soma_cluster) - 1)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        var.append(variance)

    LOGGER.info(
        f"Intracluster variance for {nb_cluster} clusters ({save_name}) : {var}"
    )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

    return var


def RepresentationOnImages(labeled_somas, kmeans, nb_cluster):
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    Represent the result of kmeans on labeled image. IN DVPT. Only available for a kmean intra-image.
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    clustered_somas = labeled_somas.copy()
    clustered_somas = nmpy.amax(clustered_somas, axis=0)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    for indx, value in enumerate(kmeans.labels_):
        for indx_axe, axe in enumerate(clustered_somas):
            for indx_pixel, pixel in enumerate(axe):
                if pixel == indx + 1:
                    clustered_somas[indx_axe][indx_pixel] = value + 1
    pl_.imshow(clustered_somas, cmap="tab20")
    pl_.title(f"n cluster = {nb_cluster}")
    pl_.show(block=True)
    pl_.close()


def _Conditions(df)->tuple[str,...]:
    """"""
    return tuple(set(df["condition"].to_numpy().tolist()))


def _Experiments(df):
    """"""
    return tuple(set(tuple(_elm) for _elm in df[["condition", "duration"]].to_numpy().tolist()))

def FeaturesStatistics(
    save_in: str = None,
    title: str = "",
    describe: bool = True,
    heatmap: bool = True,
    drop_feat: bool = True,
    distribution: bool = True,
    stat_test: bool = True,
    decision_tree: bool = True,
):
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    Return the statistics allowing the user to choose the most relevant features to feed ML algorithms for ex.
    Statistical description, correlation heatmap, dropping correlated features or features with little difference between the two conditions,
    Plotting features distribution and boxplots, performing Kolmogorov-Smirnov two-sample test and Wilcoxon signed-rank two-sample test.
    Can draw a decision tree.
DEBREUVE Eric's avatar
DEBREUVE Eric committed
    #
    # Overview of the basic stats on each columns of df
    if describe:
        description = df.describe()
        if save_in is not None:
            description.to_csv(str(path_t(save_in) / "df_stat_description.csv"))
DEBREUVE Eric's avatar
DEBREUVE Eric committed

    # Data formatting
    df_scalar = df.drop(["duration", "condition"], axis=1)
    df_cond = df.drop(["duration"], axis=1)
    df_groupby_dur = df.groupby("duration")

    # Compute the correlation matrix
    corr_matrix = df_scalar.corr().abs()

    if heatmap:
        # Plot heat map with correlation matrix btw features
        print("Heat map with correlation matrix btw features")
        # Generate a mask for the upper triangle
        mask = nmpy.triu(nmpy.ones_like(corr_matrix, dtype=nmpy.bool_))
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Set up the figure
        fig, ax = pl_.subplots(figsize=(13, 9))
        # Generate a custom diverging colormap
        cmap = sb_.diverging_palette(220, 10, as_cmap=True)
        # Draw the heatmap with the mask and correct aspect ratio
        sb_.heatmap(
            corr_matrix,
            mask=mask,
            cmap=cmap,
            center=0,
            square=True,
            linewidths=0.5,
            cbar_kws={"shrink": 0.5},
            xticklabels=False,
        )
        ax.set_title(f"Features correlation heat map{title}", fontsize=20)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        if save_in is not None:
            pl_.savefig(
                str(path_t(save_in) / f"Features correlation heat map{title}.png")
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            pl_.close()

    stat_tests_df1 = stat_tests_df2 = None
    if stat_test and (all_conditions.__len__() == 2):
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Create dictionaries to store the statistics and p-values
        dict_ks = {}
        dict_wx = {}
        dict_ks_dur = {}
        dict_wx_dur = {}

        for column in df_scalar.columns:
            # For each feature, perform a statistical test to know if the feature distribution or median is different btw CHO and DIO conditions
            cond_col_df = pd_.concat((df_scalar[column], df_cond["condition"]), axis=1)

            # Separate the conditions
            CHO = cond_col_df.loc[cond_col_df["condition"] == all_conditions[0]]
            CHO = nmpy.asarray(CHO[column])
            DIO = cond_col_df.loc[cond_col_df["condition"] == all_conditions[1]]
            DIO = nmpy.asarray(DIO[column])
DEBREUVE Eric's avatar
DEBREUVE Eric committed

            # Compare distribution between conditions (goodness of fit)
            ks = si_.stats.ks_2samp(CHO, DIO)
            dict_ks[column] = ks

            # Compare median between conditions
            wx = si_.stats.ranksums(CHO, DIO)
            dict_wx[column] = wx

            # Plot the p-values
            fig = pl_.figure(constrained_layout=False, figsize=(15, 8))
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            gs = fig.add_gridspec(ncols=2, nrows=1)
            ax1 = fig.add_subplot(gs[0, 0])
            ax2 = fig.add_subplot(gs[0, 1])
            #
            ax1.set_title(
                f"Kolmogorov-Smirnov p-value btw conditions - {column}",
                fontsize=13,
                pad=20,
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            ax1.grid()
            ax1.set_xlabel("Duration")
            ax1.set_yscale("log")
            ax1.set_ylabel("p-value")
            #
            ax2.set_title(
                f"Wilcoxon signed-rank p-value btw conditions - {column}",
                fontsize=13,
                pad=20,
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            ax2.grid()
            ax2.set_xlabel("Duration")
            ax2.set_yscale("log")
            ax2.set_ylabel("p-value")

            # Do the same thing than above but separate durations between conditions
            for duration, values in df_groupby_dur:
                if duration not in dict_ks_dur:
                    dict_ks_dur[duration] = {}
                    dict_wx_dur[duration] = {}

                duration_df = values.drop(["duration"], axis=1)
                CHO_ = duration_df.loc[duration_df["condition"] == all_conditions[0]]
                DIO_ = duration_df.loc[duration_df["condition"] == all_conditions[1]]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                CHO_ = CHO_.drop(["condition"], axis=1)
                DIO_ = DIO_.drop(["condition"], axis=1)
                CHO_ = nmpy.asarray(CHO_[column])
                DIO_ = nmpy.asarray(DIO_[column])
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                # Compare distribution
                ks2 = si_.stats.ks_2samp(CHO_, DIO_)
                dict_ks_dur[duration][column] = ks2

                # Compare median
                wx2 = si_.stats.ranksums(CHO_, DIO_)
                dict_wx_dur[duration][column] = wx2

            ax1.plot(
                [f"{duration}" for duration, _ in df_groupby_dur],
                [dict_ks_dur[duration][column][1] for duration, _ in df_groupby_dur],
                "g",
            )
            ax1.plot(
                [f"{duration}" for duration, _ in df_groupby_dur],
                [dict_ks_dur[duration][column][1] for duration, _ in df_groupby_dur],
                "ko",
            )
            ax2.plot(
                [f"{duration}" for duration, _ in df_groupby_dur],
                [dict_wx_dur[duration][column][1] for duration, _ in df_groupby_dur],
                "y",
            )
            ax2.plot(
                [f"{duration}" for duration, _ in df_groupby_dur],
                [dict_wx_dur[duration][column][1] for duration, _ in df_groupby_dur],
                "ko",
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

            pl_.tight_layout()

            if save_in:
                pl_.savefig(str(path_t(save_in) / f"p_values_duration_{column}.png"))
DEBREUVE Eric's avatar
DEBREUVE Eric committed
                pl_.close()

        # Reformat the data to save it as csv
        df_ks = pd_.DataFrame.from_dict(data=dict_ks)
        df_ks = df_ks.rename(
            index={0: "Kolmogorov-Smirnov statistic", 1: "Kolmogorov-Smirnov p-value"}
        )
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        df_wx = pd_.DataFrame.from_dict(data=dict_wx)
        df_wx = df_wx.rename(index={0: "Wilcoxon statistic", 1: "Wilcoxon p-value"})

        df_ks_dur = pd_.DataFrame()
        df_wx_dur = pd_.DataFrame()
        for key in dict_ks_dur.keys():
            df_ks_dur = pd_.concat(
                (df_ks_dur, pd_.DataFrame.from_dict(data=dict_ks_dur[key]))
            )
            df_ks_dur = df_ks_dur.rename(
                index={
                    0: f"{key} - Kolmogorov-Smirnov statistic",
                    1: f"{key} - Kolmogorov-Smirnov p-value",
                }
            )
            df_wx_dur = pd_.concat(
                (df_wx_dur, pd_.DataFrame.from_dict(data=dict_wx_dur[key]))
            )
            df_wx_dur = df_wx_dur.rename(
                index={0: f"{key} - Wilcoxon statistic", 1: f"{key} - Wilcoxon p-value"}
            )
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        stat_tests_df1 = pd_.concat((df_ks, df_wx))  # KS and Wx all durations
        stat_tests_df2 = pd_.concat(
            (df_ks_dur, df_wx_dur)
        )  # KS and Wx for each duration
DEBREUVE Eric's avatar
DEBREUVE Eric committed

        if save_in:
            stat_tests_df1.to_csv(
                str(path_t(save_in) / "stat_test_KS_wilcoxon_all.csv")
            )
            stat_tests_df2.to_csv(
                str(path_t(save_in) / "stat_test_KS_wilcoxon_for_each_duration.csv")
DEBREUVE Eric's avatar
DEBREUVE Eric committed

    if drop_feat:
        # Drop highly correlated features
        LOGGER.info("Drop highly correlated features")
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Select upper triangle of correlation matrix
        upper = corr_matrix.where(
            nmpy.triu(nmpy.ones(corr_matrix.shape), k=1).astype(nmpy.bool_)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Find index of feature columns with correlation greater than 0.9
        to_drop = [column for column in upper.columns if any(upper[column] > 0.9)]
        # Drop features
        drop_HCF_df = df_scalar.drop(df[to_drop], axis=1)

        # Drop column with null variance
        drop_HCF_df = drop_HCF_df.loc[:, drop_HCF_df.var() != 0.0]

        if stat_test and (all_conditions.__len__() == 2):
            # drop non-significant features (distribution and mean not different btw CHO and DIO)
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            drop_HCF_df_6H = drop_HCF_df.copy()
            # based on all durations;
            to_drop = [
                column
                for column in drop_HCF_df.columns
                if (stat_tests_df1.loc["Kolmogorov-Smirnov p-value", column] > 1.0e-2)
                and (stat_tests_df1.loc["Wilcoxon p-value", column] > 1.0e-2)
            ]
DEBREUVE Eric's avatar
DEBREUVE Eric committed
            drop_HCF_df = drop_HCF_df.drop(drop_HCF_df[to_drop], axis=1)

            if "6H - Kolmogorov-Smirnov p-value" in stat_tests_df2.columns:
                # only based on 6H duration;
                to_drop_6H = [
                    column
                    for column in drop_HCF_df_6H.columns
                    if (
                        stat_tests_df2.loc["6H - Kolmogorov-Smirnov p-value", column]
                        > 1.0e-3
                    )
                    and (stat_tests_df2.loc["6H - Wilcoxon p-value", column] > 1.0e-3)
                ]
                drop_HCF_df_6H = drop_HCF_df_6H.drop(drop_HCF_df_6H[to_drop_6H], axis=1)
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                drop_HCF_df_6H = pd_.concat(
                    (df[["condition", "duration"]], drop_HCF_df_6H), axis=1
DEBREUVE Eric's avatar
DEBREUVE Eric committed

                if save_in:
                    drop_HCF_df_6H.to_csv(
                        str(path_t(save_in) / "df_drop_highly_corr_feat_6H.csv")
                    )
                    print(
                        f"Selection of features with distribution and median of CHO vs DIO different in: {save_in}"
                    )

DEBREUVE Eric's avatar
DEBREUVE Eric committed
        # Add the condition and duration
        drop_HCF_df = pd_.concat((df[["condition", "duration"]], drop_HCF_df), axis=1)

        if save_in:
            drop_HCF_df.to_csv(str(path_t(save_in) / "df_drop_highly_corr_feat.csv"))