鸢尾草分类-手写简单神经网络-向量化实现

import random
from csv import reader

import matplotlib.pyplot as plt
import numpy as np
from sklearn.preprocessing import MinMaxScaler


def load_dataset(dataset_path, n_train_data):
    dataset = []
    label_dict = {'Iris-setosa': 0, 'Iris-versicolor': 1, 'Iris-virginica': 2}
    with open(dataset_path, 'r') as file:
        csv_reader = reader(file, delimiter=',')
        for row in csv_reader:
            row[0:4] = list(map(float, row[0:4]))
            row[4] = label_dict[row[4]]
            dataset.append(row)
    dataset = np.array(dataset)
    mms = MinMaxScaler()
    for i in range(dataset.shape[1] - 1):
        dataset[:, i] = mms.fit_transform(dataset[:, i].reshape(-1, 1)).flatten()
    dataset = dataset.tolist()
    for row in dataset:
        row[4] = int(row[4])
    random.shuffle(dataset)
    train_data = np.array(dataset[0:n_train_data]).T
    val_data = np.array(dataset[n_train_data:]).T
    return train_data, val_data


if __name__ == "__main__":
    file_path = './iris.csv'
    l_rate = 0.01
    epochs = 300
    # load data
    train_data, val_data = load_dataset(file_path, 130)
    y = train_data[4].reshape([1, 130])
    a0 = np.delete(train_data, 4, axis=0)  # 4*130
    w1 = np.random.random((5, 4))  # 5*4
    w2 = np.random.random((1, 5))  # 1*5
    cost = []
    for i in range(0, epochs):
        # forward
        z1 = np.dot(w1, a0)  # 5*130
        a1 = 1.0 / (1.0 + np.exp(-z1))  # 5*130
        z2 = np.dot(w2, a1)  # 1*130
        # back
        dz2 = z2 - y  # 1*130
        cost.append(np.sum(np.power(np.abs(dz2), 2)) / 130)
        dw2 = np.dot(dz2, a1.T)  # 1*5
        da1 = np.dot(w2.T, dz2)  # 5*130
        dz1 = np.multiply(da1, np.multiply(a1, 1 - a1))  # 5*130
        dw1 = np.dot(dz1, a0.T)  # 5*4
        w1 = w1 - l_rate * dw1
        w2 = w2 - l_rate * dw2
    # print("w1 =\n" + str(w1) + "\nw2 =\n" + str(w2))
    # test
    y_test = val_data[4].reshape([1, 20])
    a0_test = np.delete(val_data, 4, axis=0)
    z1_test = np.dot(w1, a0_test)
    a1_test = 1.0 / (1.0 + np.exp(-z1_test))
    z2_test = np.dot(w2, a1_test)
    z2_test = np.rint(z2_test)
    dz2_test = z2_test - y_test
    print(z2_test)
    print(y_test)
    print(np.sum(np.power(np.abs(dz2_test), 2)) / 20)
    plt.xlabel('epochs')
    plt.ylabel('cost')
    plt.plot(cost)
    plt.show()