benchmark_knn.py

# -*- coding: utf-8 -*-
"""
Benchmark script for outlier detection based on k-Nearest Neighbors (kNN).

Copyright (C) 2022 by Akira TAMAMORI
Copyright (C) 2018, Yue Zhao

Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:

* Redistributions of source code must retain the above copyright notice, this
  list of conditions and the following disclaimer.

* Redistributions in binary form must reproduce the above copyright notice,
  this list of conditions and the following disclaimer in the documentation
  and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
"""

import os
import warnings

import numpy as np
from pyod.models.knn import KNN
from scipy.io import loadmat
from sklearn.metrics import precision_recall_fscore_support, roc_auc_score
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler

import optuna

# supress warnings for clean output
warnings.filterwarnings("ignore")

NORMAL = 0
ANOMALY = 1

# split ratio of test data
TEST_SIZE = 0.4

# for Optuna
SPLIT_RATIO = 0.1

# Number of search trials for Optuna
N_TRIALS = 50

# Define the number of iterations for evaluation
N_ITER = 10

# Define data file and read X and y
MAT_FILE_LIST = [
    "arrhythmia.mat",
    "cardio.mat",
    "glass.mat",
    "ionosphere.mat",
    "letter.mat",
    "mnist.mat",
    "musk.mat",
    "optdigits.mat",
    "pendigits.mat",
    "pima.mat",
    "satellite.mat",
    "satimage-2.mat",
    "shuttle.mat",
    "vertebral.mat",
    "vowels.mat",
    "wbc.mat",
]

# benchmark scores
SCORES = {}

# temporary variables
VARS = {
    "X_train": None,
    "X_test": None,
    "y_train": None,
    "y_test": None,
    "x_train": None,
    "x_valid": None,
    "t_valid": None,
    "X_train_norm": None,
    "X_test_norm": None,
    "clf": None,
    "study": None,
    "anomaly_scores": None,
    "anomaly_threshold": None,
    "prec": 0.0,
    "recall": 0.0,
    "fscore": 0.0,
}


class Objective:
    """Objective class for Optuna."""

    def __init__(self, X_train, X_valid, y_valid, config):
        """Initialize."""
        self.X_train = X_train
        self.X_valid = X_valid
        self.y_valid = y_valid
        self.contamination = config["contamination"]

    def __call__(self, trial):
        """Call."""
        clf = KNN(
            n_neighbors=trial.suggest_int("n_neighbors", 1, 16, 1),
            leaf_size=trial.suggest_int("leaf_size", 2, 50, 2),
            contamination=self.contamination,
        )
        clf.fit(self.X_train)
        anomaly_scores = clf.decision_function(self.X_valid)
        roc = roc_auc_score(self.y_valid, anomaly_scores)

        return roc

    def print_contamination(self):
        """Dummy function to supress warnings from pylint."""
        contamination = self.contamination
        print(f"contamination={contamination:.2f}")


def print_result():
    """
    Print benchmark results.
    """

    print("\nBenchmark Results\n")
    for mfile in MAT_FILE_LIST:
        auc_ave = SCORES[mfile]["auc"]["ave"]
        auc_std = SCORES[mfile]["auc"]["std"]
        prec_ave = SCORES[mfile]["prec"]["ave"]
        prec_std = SCORES[mfile]["prec"]["std"]
        recall_ave = SCORES[mfile]["recall"]["ave"]
        recall_std = SCORES[mfile]["recall"]["std"]
        fscore_ave = SCORES[mfile]["fscore"]["ave"]
        fscore_std = SCORES[mfile]["fscore"]["std"]
        print(
            f"{mfile}: AUC={auc_ave:.4f} ± {auc_std:.4f}, "
            f"Prec={prec_ave:.4f} ± {prec_std:.4f}, "
            f"Recall={recall_ave:.4f} ± {recall_std:.4f}, "
            f"F1-score={fscore_ave:.4f} ± {fscore_std:.4f}"
        )


def benchmark():
    """
    Evaluate detectors on benchmark datasets.
    """

    for file in MAT_FILE_LIST:
        SCORES[file] = {
            "auc": {},
            "prec": {},
            "recall": {},
            "fscore": {},
        }

    for mat_file in MAT_FILE_LIST:

        mat = loadmat(os.path.join("data", mat_file))

        X = mat["X"]
        y = mat["y"].ravel()  # 0: normal, 1: anomaly

        outliers_fraction = np.count_nonzero(y) / len(y)
        scaler = StandardScaler()

        scores = {
            "auc": [],
            "prec": [],
            "recall": [],
            "fscore": [],
        }

        for i in range(N_ITER):
            print("\n... Processing", mat_file, "...", "Iteration", i + 1)
            random_state = np.random.RandomState(i)

            (
                VARS["X_train"],
                VARS["X_test"],
                VARS["y_train"],
                VARS["y_test"],
            ) = train_test_split(
                X, y, test_size=TEST_SIZE, random_state=random_state, stratify=y
            )

            # standardizing data for processing
            VARS["X_train_norm"] = scaler.fit_transform(VARS["X_train"])
            VARS["X_test_norm"] = scaler.transform(VARS["X_test"])

            # Split training data further for hyperparameter search
            (VARS["x_train"], VARS["x_valid"], _, VARS["t_valid"],) = train_test_split(
                VARS["X_train_norm"],
                VARS["y_train"],
                test_size=SPLIT_RATIO,
                stratify=VARS["y_train"],
            )

            # hyper parameter search via optuna
            VARS["study"] = optuna.create_study(direction="maximize")
            VARS["study"].optimize(
                Objective(
                    X_train=VARS["x_train"],
                    X_valid=VARS["x_valid"],
                    y_valid=VARS["t_valid"],
                    config={
                        "contamination": outliers_fraction,
                    },
                ),
                n_trials=N_TRIALS,
            )

            # re-fit with best params
            VARS["clf"] = KNN(
                n_neighbors=VARS["study"].best_params["n_neighbors"],
                leaf_size=VARS["study"].best_params["leaf_size"],
                contamination=outliers_fraction,
            )
            VARS["clf"].fit(VARS["x_train"])

            # calculate anomaly scores
            VARS["anomaly_scores"] = VARS["clf"].decision_function(VARS["X_test_norm"])
            VARS["anomaly_threshold"] = np.percentile(
                VARS["anomaly_scores"], 100 * (1 - outliers_fraction)
            )

            # calculate precision, recall, and f1-score
            (
                VARS["prec"],
                VARS["recall"],
                VARS["fscore"],
                _,
            ) = precision_recall_fscore_support(
                VARS["y_test"],
                np.where(
                    VARS["anomaly_scores"] >= VARS["anomaly_threshold"],
                    ANOMALY,
                    NORMAL,
                ),
                average="binary",
            )

            # store scores in dict
            scores["auc"].append(roc_auc_score(VARS["y_test"], VARS["anomaly_scores"]))
            scores["prec"].append(VARS["prec"])
            scores["recall"].append(VARS["recall"])
            scores["fscore"].append(VARS["fscore"])

        # calculate average and standard dev. over iterations
        for score in ("auc", "prec", "recall", "fscore"):
            SCORES[mat_file][score]["ave"] = np.average(np.array(scores[score]))
            SCORES[mat_file][score]["std"] = np.std(np.array(scores[score]))


if __name__ == "__main__":
    benchmark()
    print_result()