题解 | 手写实现的二分类预测器

import numpy as np
import sys
import json


class BinaryClassifier:
    def __init__(self, train, test):
        self.n, self.m = train.shape
        self.m -= 1
        self.t = test.shape[0]
        label = train[:, self.m]
        train = train[:, :self.m]

        self.label_zero_mask = np.argwhere(label == 0)
        self.label_one_mask = np.argwhere(label == 1)

        pi_c = self.calculate_pi() / self.n

        mu, sigma = self.character_mean_var(train)

        self.result = self.predcit(pi_c, mu, sigma, test)

        return

    def calculate_pi(self):
        pi_c = np.array([self.label_zero_mask.shape[0], self.label_one_mask.shape[0]])
        pi_c = pi_c / self.n
        return pi_c

    def character_mean_var(self, train):

        mu = np.zeros([2, self.m])
        sigma = np.zeros([2, self.m])
        for jj in range(self.m):
            train_label_zero = []
            train_label_one = []
            for ii in self.label_zero_mask:
                train_label_zero.append(train[ii, jj])
            for ii in self.label_one_mask:
                train_label_one.append(train[ii, jj])
            train_label_zero = np.array(train_label_zero)
            train_label_one = np.array(train_label_one)
            mu[0, jj] = np.mean(train_label_zero)
            mu[1, jj] = np.mean(train_label_one)
            sigma[0, jj] = np.var(train_label_zero, ddof=0)
            sigma[1, jj] = np.var(train_label_one, ddof=0)
            for cc in range(2):
                if sigma[cc, jj] == 0:
                    sigma[cc, jj] = 1e-9

        return mu, sigma

    def predcit(self, pi_c, mu, sigma, test):
        result = []
        for ii in range(self.t):
            x = test[ii]
            logP = np.log(pi_c) + np.sum(-0.5 * np.log(2 * np.pi * sigma ** 2) - (x - mu) ** 2 / (2 * sigma ** 2), axis=1)
            result.append(np.argmax(logP))
        return result


if __name__ == "__main__":
    json_input = json.load(sys.stdin)
    bincls = BinaryClassifier(np.array(json_input["train"]),
                              np.array(json_input["test"]))
    print(bincls.result)