CNN网络的维度确定（20）---《深度学习》

代码对网上mnist.py文件进行了简单的修改实现！

import os
import gzip
from urllib.request import urlopen

import numpy as np
import tensorflow as tf

def load_data(fpath='D:\\Science\\tensorflow-adversarial-master\\mnist.npz'):

    if not os.path.exists(fpath):
        print('Downloading mnist dataset')
        _mkdata(fpath)

    db = np.load(fpath)
    X_train, y_train = db['x_train'], db['y_train']
    X_test, y_test = db['x_test'], db['y_test']
    y_train=to_categorial(y_train,10)
    y_test=to_categorial(y_test,10)
    X_train=np.reshape(X_train,(X_train.shape[0], X_train.shape[1]*X_train.shape[2]))
    X_test=np.reshape(X_test,(X_test.shape[0], X_test.shape[1]*X_test.shape[2]))
    return (X_train, y_train), (X_test, y_test)


def _mkdata(fpath):
    _download('/var/tmp/mnist_X_train.gz',
              'http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz')
    _download('/var/tmp/mnist_y_train.gz',
              'http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz')
    _download('/var/tmp/mnist_X_test.gz',
              'http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz')
    _download('/var/tmp/mnist_y_test.gz',
              'http://yann.lecun.com/exdb/mnist/t10k-labels-idx1-ubyte.gz')

    X_train = _extract_images('/var/tmp/mnist_X_train.gz')
    y_train = _extract_labels('/var/tmp/mnist_y_train.gz')
    y_train = np.expand_dims(y_train, 1)
    X_test = _extract_images('/var/tmp/mnist_X_test.gz')
    y_test = _extract_labels('/var/tmp/mnist_y_test.gz')
    y_test = np.expand_dims(y_test, 1)
    np.savez_compressed(fpath, X_train=X_train, y_train=y_train,
                        X_test=X_test, y_test=y_test)


def _read32(bytestream):
    dt = np.dtype(np.uint32).newbyteorder('>')
    return np.frombuffer(bytestream.read(4), dtype=dt)[0]


def _extract_images(fpath):
    with gzip.GzipFile(fpath) as bytestream:
        magic = _read32(bytestream)
        if magic != 2051:
            raise ValueError('Invalid magic number {0} in MNIST image file: {1}'.format(magic, f.name))
        num_images = _read32(bytestream)
        rows = _read32(bytestream)
        cols = _read32(bytestream)
        buf = bytestream.read(rows*cols*num_images)
        data = np.frombuffer(buf, dtype=np.uint8)
        data = data.reshape(num_images, rows, cols)
    return data


def _extract_labels(fpath):
    with gzip.GzipFile(fpath) as bytestream:
        magic = _read32(bytestream)
        if magic != 2049:
            raise ValueError('Invalid magic number {0} in MNIST label file: {1}'.format(magic, f.name))
        num_items = _read32(bytestream)
        buf = bytestream.read(num_items)
        labels = np.frombuffer(buf, dtype=np.uint8)
        return labels


def _download(fpath, url):
    if os.path.exists(fpath):
        return
    workdir = os.path.dirname(fpath) or '.'
    os.makedirs(workdir, exist_ok=True)
    with urlopen(url) as ret, open(fpath, 'wb') as w:
        w.write(ret.read())
def to_categorial(y,n_classes):
    y_std=np.zeros([len(y),n_classes])
    for i in range(len(y)):
        y_std[i,y[i]]=1.0
    return y_std

if __name__ == '__main__':
    #load_data()
    (X_train, y_train), (X_test, y_test) = load_data()
    x=tf.placeholder(tf.float32,[None,784])
    W=tf.Variable(tf.zeros([784,10]))
    b=tf.Variable(tf.zeros([10]))
    y=tf.nn.softmax(tf.matmul(x,W)+b)

    y_=tf.placeholder(tf.float32,[None,10])
    z=y_*tf.log(y)
    cross=tf.reduce_sum(y_*tf.log(y),[1])
    cross_entropy=tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
    train_step = tf.train.GradientDescentOptimizer(0.5).minimize(cross_entropy)

    correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1))
    accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))


    init_op=tf.global_variables_initializer()
    with tf.Session() as sess:
        sess.run(init_op)
        batch_xs, batch_ys = X_train[0:200],y_train[0:200]
        #sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})

        print(sess.run(cross_entropy,feed_dict={x: batch_xs, y_: batch_ys}))

1）我们来确定z的维度，即y_*tf.log(y)返回的维度：

z=y_*tf.log(y)
...
sess.run((z, feed_dict={x: batch_xs, y_: batch_ys}).shape)

运行结果：

所以我们可以确定这里的”*”是点乘运算，而不是矩阵乘法，即（N,10）和（N,10）的矩阵想乘仍旧是（N,10），因为对应的每个位置进行相应的点乘运算即可！

2）现在我们来确定cross的维度，即tf.reduce_sum(y_*tf.log(y),[1])的维度：

cross=tf.reduce_sum(y_*tf.log(y),[1])
...
print(sess.run(cross,feed_dict={x: batch_xs, y_: batch_ys}).shape)

运行结果：

我们也可以来看一下其中的值：

cross=tf.reduce_sum(y_*tf.log(y),[1])
...
print(sess.run(cross,feed_dict={x: batch_xs, y_: batch_ys}))

运行结果：

3）现在来看cross_entropy的值：

cross_entropy=tf.reduce_mean(-tf.reduce_sum(y_*tf.log(y),reduction_indices=[1]))
...
print(sess.run(cross_entropy,feed_dict={x: batch_xs, y_: batch_ys}))

运行结果：