多元线性回归(用numpy和tensorflow分别实现)

...以前所有机器的学习代码都是看别人写的, 这个算是第一个自己独立写出的代码吧(虽然很简单

加油~~~

import numpy as np
import tensorflow as tf
import time

# generate data
n_dims = 100
x = np.random.randn(10000, n_dims)
w1 = np.random.rand(n_dims)
b1 = [-0.32]
y = np.matmul(x, w1) + b1

# init: np和tf共享w,b
w = np.random.randn(n_dims)
b = np.random.randn(1)
l_t = 0.00001  # 学习率
epoch = 1000


# np_train过程
def train(x, targt, w, b):
    y = np.matmul(x, w) + b
    loss = np.sum((y - targt) ** 2)
    z = x * (y - targt)[:, np.newaxis]
    w_g = np.sum(z, axis=0)
    b_g = np.sum(y - targt)
    w -= l_t * w_g
    b -= l_t * b_g
    return loss, w, b


# np
def nmp(x, y, w, b):
    t1 = time.time()
    for i in range(epoch):
        loss, w, b = train(x, y, w, b)
        if i % 100 == 0:
            print(i, loss)
    print('np_time:', time.time() - t1)
    return w, b


# tf
def tfw(x, y, w, b):
    t1 = time.time()
    wf = tf.Variable(w, dtype='float')
    bf = tf.Variable(b, dtype='float')
    xf = tf.placeholder('float', [None, n_dims])
    yf = tf.placeholder('float', [None, 1])
    # wf = tf.expand_dims(wf, -1)
    out = tf.add(tf.matmul(xf, wf), bf)
    # tf的矩阵乘法需要二维!!!!!!!!!
    loss = tf.reduce_sum(tf.pow(out - yf, 2))
    optimizer = tf.train.GradientDescentOptimizer(l_t).minimize(loss)
    init = tf.global_variables_initializer()
    with tf.Session() as sess:
        sess.run(init)
        y = y[:, np.newaxis]  # 扩展y为二维矩阵
        for i in range(epoch):
            _loss, _ = sess.run([loss, optimizer], feed_dict={xf: x, yf: y})
            if i % 100 == 0:
                print(i, _loss)
        print("tf_time:", time.time() - t1)
        return sess.run([wf, bf])


# test
tmp = np.random.rand(n_dims)
ans = np.sum(tmp * w1) + b1
# np
wn, bn = w, b
wn, bn = nmp(x, y, wn, bn)
print(ans - np.sum(tmp * wn) - bn)  #: time:11S, error: 1e-25
# tf
w = w[:, np.newaxis]
wf, bf = tfw(x, y, w, b)
print(ans - np.sum(tmp * wf.reshape(-1)) - bf)  #: time:6S, error: 4e-9