TensorFlow逻辑运算（（A == B）&&（C == D））导致“形状不兼容：[2]与[3]”

Question

I'm trying to build the following logical expression -

tf.logical_and(tf.equal(tf.argmax(y_conv,0), tf.argmax(y_,0)), 
                tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1)), name=None)

But it results in the below error -

Incompatible shapes: [2] vs. [3]

tf.equal(tf.argmax(y_conv,0), tf.argmax(y_,0)) and tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1)) work fine separately, the error occurs only with tf.logical_and . tf.logical_and expects boolean tensors and tf.equal returns boolean tensors, so all arguments are in order so not sure why it fails.

To give some context, the original code is below and I'm just trying to update correct_prediction to include both 0 and 1 for tf.argmax

UPDATE1 Start (this adds all of the variable declarations)

sess = tf.InteractiveSession()
def weight_variable(shape):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial)
def bias_variable(shape):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial)
def conv2d(x, W): return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
def max_pool_2x2(x): return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME')
W_conv1 = weight_variable([3, 3, 1, 32])
b_conv1 = bias_variable([32])
x = tf.placeholder(tf.float32, shape=[None, 9])
x_image = tf.reshape(x, [-1,3,3,1])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
h_pool1 = max_pool_2x2(h_conv1)
h_pool2_flat = tf.reshape(h_conv1, [-1, 3 * 3 * 32])
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
W_fc2 = weight_variable([64, 2])
b_fc2 = bias_variable([2])
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)

UPDATE1 End (this adds all of the variable declarations)

This is where the problem is located -

y_ = tf.placeholder(tf.float32, shape=[None, 2])
y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
#This works - correct_prediction = tf.equal(tf.argmax(y_conv,0), tf.argmax(y_,0)) . Changed it to -
correct_prediction = tf.logical_and(tf.equal(tf.argmax(y_conv,0), tf.argmax(y_,0)), 
                tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1)), 
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
sess.run(tf.global_variables_initializer())
train_step.run(feed_dict={x: xtrain, y_: ytrain, keep_prob: 0.5})
#In debugging mode, code breaks at the below line
print("test accuracy %g"%accuracy.eval(feed_dict={x: xtest, y_: ytest, keep_prob: 1.0}))

How can I debug this error?

Answer 1

The problem arises because tf.equal() is an elementwise operation and it returns a tensor with the same shape as its arguments. The easiest way to fix your expression is to use tf.reduce_all() to aggregate the results of tf.equal() down to a scalar before computing the tf.logical_and() , as follows:

tf.logical_and(
    tf.reduce_all(tf.equal(tf.argmax(y_conv, 0), tf.argmax(y_, 0))), 
    tf.reduce_all(tf.equal(tf.argmax(y_conv,1), tf.argmax(y_,1))))