【机器学习TensorFlow】【手写识别】【全连接】请大家看下为什么权重没有更新

Question

今天尝试根据网上教程写了手写识别的实例，发现打印权重weights的时候，随着迭代次数权重不会改变。但其他变量如bias是随迭代而改变的。请问大家这个权重为什么不会更新呢？

import tensorflow as tf
import input_data

tf.compat.v1.disable_eager_execution()


def full_connection():
    # 1. 准备数据
    mnist = input_data.read_data_sets("./mnist_data", one_hot=True)
    x = tf.compat.v1.placeholder(dtype=tf.float32, shape=(None, 784))
    y_true = tf.compat.v1.placeholder(dtype=tf.float32, shape=(None, 10))

    # 2. 构建模型
    weights = tf.compat.v1.Variable(initial_value=tf.compat.v1.random_normal(shape=[784, 10]), name="weights")
    bias = tf.compat.v1.Variable(initial_value=tf.compat.v1.random_normal(shape=[10]), name="bias")
    y_predict = tf.matmul(x, weights) + bias

    # 3. 构造损失函数
    error = tf.nn.softmax_cross_entropy_with_logits(labels=y_true, logits=y_predict)
    error = tf.reduce_mean(error)

    # 4. 优化损失
    optimizer = tf.compat.v1.train.GradientDescentOptimizer(learning_rate=0.03).minimize(error)

    # 5. 准确率计算
    equal_list = tf.equal(tf.argmax(y_true, 1), tf.argmax(y_predict, 1))
    accuracy = tf.reduce_mean(tf.cast(equal_list, tf.float32))

    # 6. 增加变量显示
    tf.compat.v1.summary.scalar("error", error)
    tf.compat.v1.summary.histogram("weights", weights)
    tf.compat.v1.summary.histogram("bias", bias)
    merged = tf.compat.v1.summary.merge_all()
    saver = tf.compat.v1.train.Saver()

    with tf.compat.v1.Session() as sess:
        init = tf.compat.v1.global_variables_initializer()
        sess.run(init)
        image, label = mnist.train.next_batch(100)
        file_writer = tf.compat.v1.summary.FileWriter("./log", graph=sess.graph)

        for i in range(1000):
            _, loss, accuracy_value = sess.run([optimizer, error, accuracy], feed_dict={x: image, y_true: label})
            print("第%d次训练，损失为%f，准确率为%f" % (i+1, loss, accuracy_value))
            print(weights.eval())
            summary_result = sess.run(merged, feed_dict={x: image, y_true: label})
            file_writer.add_summary(summary_result, i)
            if i % 10 ==0:
                saver.save(sess, "./log/model/model-1.ckpt")

    return None


if __name__ == "__main__":
    full_connection()

第6次训练，损失为16.070484，准确率为0.090000
权重为：
 [[-6.5291589e-01 -1.6071169e+00  2.1290806e-01 ... -2.1784568e+00
  -1.2091833e+00 -8.6140698e-01]
 [ 1.2597474e+00 -1.4910623e-01 -3.8707003e-01 ...  1.0214819e-01
   5.9081298e-01 -6.1707306e-01]
 [-1.4509798e+00 -1.9147521e-02  1.6300718e+00 ... -2.3356327e-01
   2.2199700e+00  1.0290977e+00]
 ...
 [ 8.4160960e-01 -4.5380855e-01  4.1801175e-01 ...  3.1385022e-01
  -1.0918922e+00  1.6047851e+00]
 [ 1.5874079e-01  1.3555746e+00  7.1338147e-01 ...  5.6245077e-01
   1.2595990e+00 -8.3930290e-01]
 [ 7.4030966e-01  2.1435671e+00 -4.2898212e-02 ... -5.6554389e-01
   7.3194194e-01 -7.8909863e-05]]


第112次训练，损失为6.315756，准确率为0.270000
权重为：
 [[-6.5291589e-01 -1.6071169e+00  2.1290806e-01 ... -2.1784568e+00
  -1.2091833e+00 -8.6140698e-01]
 [ 1.2597474e+00 -1.4910623e-01 -3.8707003e-01 ...  1.0214819e-01
   5.9081298e-01 -6.1707306e-01]
 [-1.4509798e+00 -1.9147521e-02  1.6300718e+00 ... -2.3356327e-01
   2.2199700e+00  1.0290977e+00]
 ...
 [ 8.4160960e-01 -4.5380855e-01  4.1801175e-01 ...  3.1385022e-01
  -1.0918922e+00  1.6047851e+00]
 [ 1.5874079e-01  1.3555746e+00  7.1338147e-01 ...  5.6245077e-01
   1.2595990e+00 -8.3930290e-01]
 [ 7.4030966e-01  2.1435671e+00 -4.2898212e-02 ... -5.6554389e-01
   7.3194194e-01 -7.8909863e-05]]

 [[-6.5291589e-01 -1.6071169e+00  2.1290806e-01 ... -2.1784568e+00
  -1.2091833e+00 -8.6140698e-01]
 [ 1.2597474e+00 -1.4910623e-01 -3.8707003e-01 ...  1.0214819e-01
   5.9081298e-01 -6.1707306e-01]
 [-1.4509798e+00 -1.9147521e-02  1.6300718e+00 ... -2.3356327e-01
   2.2199700e+00  1.0290977e+00]
 ...
 [ 8.4160960e-01 -4.5380855e-01  4.1801175e-01 ...  3.1385022e-01
  -1.0918922e+00  1.6047851e+00]
 [ 1.5874079e-01  1.3555746e+00  7.1338147e-01 ...  5.6245077e-01
   1.2595990e+00 -8.3930290e-01]
 [ 7.4030966e-01  2.1435671e+00 -4.2898212e-02 ... -5.6554389e-01
   7.3194194e-01 -7.8909863e-05]]

Answer 1

你好，我是有问必答小助手，非常抱歉，本次您提出的有问必答问题，技术专家团超时未为您做出解答


本次提问扣除的有问必答次数，将会以问答VIP体验卡（1次有问必答机会、商城购买实体图书享受95折优惠）的形式为您补发到账户。


因为有问必答VIP体验卡有效期仅有1天，您在需要使用的时候【私信】联系我，我会为您补发。