如何从tensorflow einsum结果中获取numpy数组？

Question

y = tf.einsum('aij,jk->aik', x, W) + b

y.shape returns the following:

tf.Tensor 'text-representation/add:0' shape=(?, 80, 256) dtype=float32

How can I get the 80x256 array into the numpy array? I am new to tensorflow and struggling. Thanks.

Answer 1

They are two ways to approach this:

1). Execute the Tensor within a Session.

y = tf.einsum('aij,jk->aik', x, W) + b

sess = tf.Session()
print(sess.run(y))

# store the numpy array
my_arr = sess.run(y)
my_arr.shape

sess.close()  # close the session object

2). Use Eager Execution for immediate execution.

# enable eager execution
tf.enable_eager_execution()

y = tf.einsum('aij,jk->aik', x, W) + b
print(y)

Note, however, that to use Eager Execution it must be enabled before any other operation is carried out on the terminal. Moreso, when Eager Execution is enabled, it cannot be disabled unless the terminal is restarted.