analysis_utils.py

import argparse
import csv
import glob
import os
import re
from typing import List

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
from sqlalchemy import asc
from sqlalchemy.exc import OperationalError

from package.orm_models import Result, local_create_session


def argument_parser(args) -> argparse.Namespace:
    """Parse input arguments."""
    parser = argparse.ArgumentParser(
        description='''Analysis over input data''')
    parser.add_argument('--input-file', type=str, help='''input CSV file''', default=None)
    parser.add_argument('-o', '--output_file', type=str, help='''output file''')
    parser.add_argument('-p', '--plot-correlation', action='store_true', help='plot correlation')
    parser.add_argument('--axis1', type=str, help='''first plot axis''', default='silhouette_sklearn')
    parser.add_argument('--axis2', type=str, help='''second plot axis''', default='adjusted_rand_score')
    parser.add_argument('--color', type=str, help='''point color''', default='generation')
    parser.add_argument('--correlation', action='store_true', help='generates correlation matrix')
    parser.add_argument('--merge-results', action='store_true', help='')
    parser.add_argument('--n-last-results', type=int, help='', default=10)
    parser.add_argument('--useful-features', action='store_true', help='')
    parser.add_argument('--clear-incomplete-outputs', action='store_true', help='')
    parser.add_argument('--melt-results', action='store_true', help='')
    parser.add_argument('--average-feature-selection', action='store_true', help='')
    parser.add_argument('--list-results', action='store_true')
    parser.add_argument('--detail-result', action='store_true')
    parser.add_argument('--confusion-matrix', action='store_true')
    parser.add_argument('--filter', action='store_true')
    parser.add_argument('--id', type=int, default=None, nargs='+')
    parser.add_argument('--exp-name', type=str, default=None)
    parser.add_argument('--db-file', type=str, default='./local.db', help='sqlite file to store results')
    parser.add_argument('--select-best', action='store_true')
    parser.add_argument('--logs-folder', type=str, default=None)
    parser.add_argument('--plot-logs', action='store_true')
    parser.add_argument('--export-results', action='store_true')
    parser.add_argument('--export-history', action='store_true')

    args = parser.parse_args(args=args)

    if sum([args.plot_correlation, args.correlation,
            args.merge_results,
            args.useful_features, args.melt_results,
            args.average_feature_selection, args.list_results,
            args.detail_result, args.confusion_matrix,
            args.filter, args.select_best,
            args.plot_logs, args.export_results,
            args.export_history]) != 1:
        raise ValueError("Cannot have this combination of arguments.")

    return args


def confusion_matrix(args):
    if args.id is None and args.exp_name is None:
        raise ValueError(f'Both id and exp-name attributes cannot be empty')

    session = local_create_session(args.db_file)

    if args.id:
        try:
            results: List[Result] = session.query(Result).filter(Result.id.in_(args.id)).all()

        except OperationalError:
            print(f'No results found with id {args.id}')
    else:
        try:
            results: List[Result] = session.query(Result).filter(Result.name == args.exp_name).all()

        except OperationalError:
            print(f'No results found with name {args.exp_name}')

    cm = sum([result.confusion_matrix.as_dataframe() for result in results])

    if args.output_file:
        cm.to_csv(args.output_file, quoting=csv.QUOTE_NONNUMERIC, float_format='%.10f', index=True)
    # else:
    #     with pd.option_context('display.max_rows', None, 'display.max_columns', None):
    #         print(cm)

    session.close()

    return cm


def plot_correlation(db_file, axis1, axis2, color, id=None, exp_name=None):
    if id is None and exp_name is None:
        raise ValueError(f'Both id and exp-name attributes cannot be empty')

    session = local_create_session(db_file)

    individual_evaluations = []
    if id:
        try:
            results: List[Result] = session.query(Result).filter(Result.id.in_(id)).all()
            if len(results) == 0:
                raise ValueError(f'No results found with name {exp_name}')

            if len(results) == 1:
                individual_evaluations = results[0].individual_evaluations
            else:
                individual_evaluations = pd.concat([result.individual_evaluations for result in results])

        except OperationalError:
            print(f'No results found with id {id}')
    else:
        try:
            results: List[Result] = session.query(Result).filter(Result.name == exp_name).all()
            if len(results) == 0:
                raise ValueError(f'No results found with name {exp_name}')

            individual_evaluations = pd.concat([result.individual_evaluations for result in results])

        except OperationalError:
            print(f'No results found with name {exp_name}')

    df = individual_evaluations

    if type(df) == str:
        return

    df.sort_values(by='generation', ascending=True)
    # df = df.sample(frac=1 / len(results))

    plt.figure()
    x = df[axis1].values
    y = df[axis2].values
    c = df[color].values
    points = plt.scatter(x, y, c=c, s=1, cmap='viridis', alpha=0.5)
    plt.colorbar(points, label=color)
    sns.regplot(axis1, axis2, data=df, scatter=False, x_jitter=0.005, y_jitter=0.005, order=1, robust=False)

    plt.show()

    session.close()


def melt_results(args):
    summary_files = glob.glob(args.input_file + '/dataset_analysis*_metrics.csv')
    summary_files.sort(key=os.path.getmtime, reverse=True)
    print('Found ' + str(len(summary_files)) + ' files')
    summary_files = summary_files[:min(len(summary_files), args.n_last_results)]
    print('Processing ' + str(len(summary_files)) + ' files:')

    dfs = [pd.read_csv(summary, index_col=0) for summary in summary_files]
    for i, df in enumerate(dfs):
        df['trial'] = pd.Series([i] * df.shape[0])

    df = pd.concat(dfs)
    df = df.reset_index()
    df['index'] = df['index'].apply(lambda x: round(float(x) / 128))

    _, ax = plt.subplots()

    for key, grp in df.groupby(['trial', 'index']):
        print(grp)
        ax = grp.mean().plot(ax=ax, kind='line', x='index', y='silhouette_sklearn', c=key, label=key)

    plt.legend(loc='best')

    if args.output_file:
        plt.savefig(args.output_file, dpi=600)
    else:
        plt.show()


def correlation_calculation(df, args):
    correlation = df.corr()

    if args.output_file:
        correlation.to_csv(args.output_file, quoting=csv.QUOTE_NONNUMERIC, float_format='%.10f', index=True)
    else:
        print(correlation)


def merge_results(args):
    session = local_create_session(args.db_file)

    results: List[Result] = session.query(Result).filter(Result.name == args.exp_name).all()

    all_results = [[result.accuracy, result.f_measure, result.silhouette] for result in results]
    all_results = pd.DataFrame(all_results, columns=['accuracy', 'f_measure', 'silhouette'])
    all_results = pd.concat([all_results, all_results.describe()])

    session.close()

    if args.output_file:
        all_results.to_csv(args.output_file, quoting=csv.QUOTE_NONNUMERIC, float_format='%.10f', index=True)
    else:
        print(all_results)


def plot_logs(args):
    raise NotImplementedError()


def correct_wells_names(well_name):
    if well_name == 'ENC-1A-RJS':
        well_name = '1-ENC-0001A-RJS'
    if well_name == 'MRK-5-RJS':
        well_name = '1-MRK-0005-RJS'
    if well_name == 'MRK-4P-RJS':
        well_name = '1-MRK-0004P-RJS'
    if well_name == 'RJS-100':
        well_name = '1-RJS-0100-RJ'
    if well_name == 'RJS-105':
        well_name = '1-RJS-0105-RJ'
    if well_name == 'ENC-2-RJS':
        well_name = '3-ENC-0002-RJS'
    if well_name == 'ENC-3-RJS':
        well_name = '3-ENC-0003-RJS'
    if well_name == 'PRG-1-RJS':
        well_name = '3-PRG-0001-RJS'

    return well_name


def find_useful_features(args):
    df = pd.read_csv(args.input_file, index_col=0)
    del df['petrofacie']

    full_std = df.std()

    groups_std = {}
    for key, group in df.groupby('predicted labels'):
        groups_std[key] = full_std / group.std()
        for column in group:
            if group[column].apply(lambda x: x == 0).all():
                groups_std[key][column] = pd.Series([0])

    df = pd.DataFrame.from_dict(groups_std)
    df['full std'] = full_std
    df.sort_values(by=['full std'])

    if args.output_file:
        df.to_csv(args.output_file, quoting=csv.QUOTE_NONNUMERIC, float_format='%.10f', index=True)
    else:
        print(df)


def average_feature_selection(args):
    session = local_create_session(args.db_file)

    results: List[Result] = session.query(Result).filter(Result.name == args.exp_name).all()
    avg_n_features = np.mean([result.final_n_features for result in results])

    print(f'Average number of features: {avg_n_features}')

    session.close()


def show_results(args):
    session = local_create_session(args.db_file)

    try:
        results = session.query(Result).order_by(asc(Result.start_time)).all()

        for result in results:
            print(result)

    except OperationalError:
        print('No results found')

    session.close()


def show_result(args):
    session = local_create_session(args.db_file)

    try:
        result = session.query(Result).filter(Result.id == args.id).first()
        print(result.details())
    except AttributeError:
        print(f'Result {args.id} not found in {args.db_file}')

    session.close()


def filter_dataset(db_file):

    session = local_create_session(db_file)

    results = session.query(Result).all()
    for result in results:
        columns = [selected_feature.column for selected_feature in result.selected_features]
        acc = float(result.scores_to_dict()["accuracy"])
        scenario = result.args_to_dict()["scenario"]
        n_final = len(columns)
        # score = n_final/n_init
        print(f'[{result.id}]{result.name} {scenario} - {acc} {n_final}')
    session.close()


def select_best(args):
    session = local_create_session(args.db_file)

    results: List[Result] = session.query(Result).filter(Result.name == args.exp_name).all()

    df = [[result.id, result.accuracy, result.f_measure, result.silhouette] for result in results]
    df = pd.DataFrame(df, columns=['id', 'accuracy', 'f_measure', 'silhouette'])

    best_accuracy = df.iloc[df['accuracy'].idxmax()]
    print(f'''Best Accuracy: {best_accuracy}''')

    best_f_measure = df.iloc[df['f_measure'].idxmax()]
    print(f'''Best F-Measure: {best_f_measure}''')

    best_silhouette = df.iloc[df['silhouette'].idxmax()]
    print(f'''Best Silhouette: {best_silhouette}''')

    session.close()


def export_results(args):
    """In case of duplicate fitness metrics for the same dataset, the last execution is selected"""
    session = local_create_session(args.db_file)
    results = session.query(Result).all()

    experiment_names = [result.args_to_dict()['experiment_name'] for result in results]
    scenarios_names = [result.args_to_dict()['scenario'] for result in results]

    index = pd.MultiIndex.from_arrays([experiment_names, scenarios_names], names=['dataset', 'scenario'])

    df_rows = []
    for result in results:
        scores_dict = result.scores_to_dict()
        data = [[float(val) for val in scores_dict.values()]]
        df_rows += [pd.DataFrame(data=data, columns=scores_dict.keys())]

    df: pd.DataFrame = pd.concat(df_rows)
    df.index = index
    df.sort_index(axis=0, level=0, inplace=True)
    df = df.groupby(level=[0, 1]).mean()

    writer = pd.ExcelWriter(args.output_file)
    df.to_excel(writer)
    writer.close()

    session.close()


def export_history(args):
    session = local_create_session(args.db_file)
    results = session.query(Result).all()

    writer = pd.ExcelWriter(args.output_file)
    for i, result in enumerate(results):
        sheet_df = pd.DataFrame(columns=['accuracy'])
        args = result.args_to_dict()
        sheet_name = re.sub('[^a-zA-Z,]', '', result.name+args['scenario'])
        sheet_name = str(i) + sheet_name

        for gen, generation_group in result.individual_evaluations.groupby(['generation'], sort=False):
            sheet_df.loc[gen] = [max(generation_group['accuracy'])[0]]

        sheet_df.to_excel(excel_writer=writer, sheet_name=sheet_name[:30])
    writer.close()

    session.close()


def main(args=None):
    args = argument_parser(args)

    result = None

    if args.plot_correlation:
        plot_correlation(args.db_file, args.axis1, args.axis2, args.color, id=args.id, exp_name=args.exp_name)
    elif args.correlation:
        df = pd.read_csv(args.input_file, index_col=0)
        correlation_calculation(df, args)
    elif args.merge_results:
        merge_results(args)
    elif args.useful_features:
        find_useful_features(args)
    elif args.melt_results:
        melt_results(args)
    elif args.average_feature_selection:
        average_feature_selection(args)
    elif args.list_results:
        show_results(args)
    elif args.detail_result:
        show_result(args)
    elif args.confusion_matrix:
        result = confusion_matrix(args)
    elif args.filter:
        filter_dataset(args.db_file)
    elif args.select_best:
        select_best(args)
    elif args.plot_logs:
        plot_logs(args)
    elif args.export_results:
        export_results(args)
    elif args.export_history:
        export_history(args)

    return result


if __name__ == '__main__':
    main()