Tensorflow错误：形状必须为0级，但对于'cond_1 / Switch'为1级

Question

I am new to tensorflow and I am following some online exercises to get familiar with tensorflow. I want to do the following task:

Create two tensors x and y of shape 300 from any normal distribution. Use tf.cond() to return:

• The mean squared error of (x - y) , if the average of all elements in (x - y) is negative.

• The sum of absolute value of all elements in the tensor (x - y) otherwise.

My implementation:

x = tf.random_normal([300])
y = tf.random_normal([300])
mse = lambda: tf.losses.mean_squared_error(y, x)
absval = lambda: tf.abs(tf.subtract(x, y))
out = tf.cond(tf.less(x, y), mse, absval)

Error:

Shape must be rank 0 but is rank 1 for 'cond_1/Switch' (op: 'Switch') with input shapes: [300], [300]

Answer 1

Try this:

x = tf.random_normal([300])
y = tf.random_normal([300])
mse = lambda: tf.losses.mean_squared_error(y, x)
absval = lambda: tf.reduce_sum(tf.abs(tf.subtract(x, y)))
out = tf.cond(tf.reduce_mean(x - y) < 0, mse, absval)