PrivacyScanData/ML/Machine-Learining-Security/深度学习算法实现/rnn_demo.py

import numpy as np
from model.rnn import RecurrentLayer
from model import IdentityActivator, ReluActivator


def data_set():
    x = [np.array([[1], [2], [3]]),
         np.array([[2], [3], [4]])]
    d = np.array([[1], [2]])
    return x, d


def gradient_check():
    """ 梯度检查 """
    # 设计一个误差函数，取所有节点输出项之和
    error_function = lambda o: o.sum()

    rl = RecurrentLayer(3, 2, IdentityActivator(), 1e-3)

    # 计算forward值
    x, d = data_set()
    rl.forward(x[0])
    rl.forward(x[1])

    # 求取sensitivity map
    sensitivity_array = np.ones(rl.state_list[-1].shape, dtype=np.float64)
    # 计算梯度
    rl.backward(sensitivity_array, IdentityActivator())
    # 检查梯度
    epsilon = 10e-4
    for i in range(rl.W.shape[0]):
        for j in range(rl.W.shape[1]):
            rl.W[i, j] += epsilon
            rl.reset_state()
            rl.forward(x[0])
            rl.forward(x[1])
            err1 = error_function(rl.state_list[-1])
            rl.W[i, j] -= 2 * epsilon
            rl.reset_state()
            rl.forward(x[0])
            rl.forward(x[1])
            err2 = error_function(rl.state_list[-1])
            expect_grad = (err1 - err2) / (2 * epsilon)
            rl.W[i, j] += epsilon
            print('weights(%d,%d): expected - actural %f - %f' % (i, j, expect_grad, rl.gradient[i, j]))


def test():
    l = RecurrentLayer(3, 2, ReluActivator(), 1e-3)
    x, d = data_set()
    l.forward(x[0])
    l.forward(x[1])
    l.backward(d, ReluActivator())
    return x, d, l


if __name__ == '__main__':
    # gradient_check()
    a, b, rnn = test()
    print(a)
    print(rnn.gradient_list)