python - Tensorflow MNIST 准确度计算不正确

Question

我刚刚开始使用 tensorflow ，我正在尝试使用具有 2 个隐藏层和一个带有 softmax 函数的输出层的神经网络对 MNIST 数据集中的图像进行分类。我使用 minibatch gd 进行优化，并跟踪每个时期后最后一个 minibatch 的准确性。

def fetch_batch(batch_index, batch_size, data=train_data, labels=train_labels):
    low_ind = batch_index*batch_size
    upp_ind = (batch_index+1)*batch_size
    if upp_ind < data.shape[0]:
        return data[low_ind:upp_ind], labels[low_ind:upp_ind]
    else:
        return data[low_ind:], labels[low_ind:]

n_inputs = 28*28 # MNIST image size
n_hidden_1 = 300
n_hidden_2 = 100
n_outputs = 10 # ten different classes

learning_rate = 0.01

X = tf.placeholder(tf.float32, shape=(None, n_inputs), name="X")
y = tf.placeholder(tf.int64, shape=(None), name="y")

with tf.name_scope("dnn"):
    hidden_1 = tf.layers.dense(X, n_hidden_1, name="hidden_1", activation=tf.nn.relu)
    hidden_2 = tf.layers.dense(hidden_1, n_hidden_2, name="hidden_2", activation=tf.nn.relu)
    logits = tf.layers.dense(hidden_2, n_outputs, name="outputs")

with tf.name_scope("loss"):
    xentropy = tf.nn.sparse_softmax_cross_entropy_with_logits(labels=y, logits=logits)
    loss = tf.reduce_mean(xentropy, name="loss")

with tf.name_scope("train"):
    optimizer = tf.train.GradientDescentOptimizer(learning_rate)
    training_op = optimizer.minimize(loss)

with tf.name_scope("eval"):
    correct = tf.nn.in_top_k(logits, y, 1)
    accuracy = tf.reduce_mean(tf.cast(correct, tf.float32))

init = tf.global_variables_initializer()
saver = tf.train.Saver()

batch_size = 50
n_epochs = 50
m = train_data.shape[0]

with tf.Session() as sess:
    init.run()
    for epoch in range(n_epochs):
        for batch_index in range(m//batch_size):
            X_minibatch, y_minibatch = fetch_batch(batch_index, batch_size)
            #X_batch, y_batch = mnist.train.next_batch(batch_size)
            sess.run(training_op, feed_dict={X: X_minibatch, y: y_minibatch})
        acc_train = accuracy.eval(feed_dict={X: X_minibatch, y: y_minibatch})
        acc_val = accuracy.eval(feed_dict={X: mnist.validation.images, y: mnist.validation.labels})
        print(epoch, "Train accuracy: ", acc_train, " Val accuracy: ", acc_val)

当使用 MNIST 帮助程序进行训练时，我获得了正确的精度(我用于验证精度的精度)，但是我想知道为什么我自己的实现不起作用，因为它总是输出 0.0 的精度。我的数据中的小批量形状和 tensorflow 助手中的小批量形状是相同的。提前致谢!

Answer 1

您需要标准化您的数据，例如

train_data = train_data / 255.0
validation_data = validation_data / 255.0

如果您在谷歌上搜索“为什么我应该标准化机器学习中的数据”，您就会发现它为何如此重要。