Unpack tensorflow Tensor columns into separate variables

Question

Given a tensor of the form

T = tf.placeholder(tf.float32, [None, 2])

How can I separate T into two separate tensors of shape (?, 1), one for each column?

Answer 1

You can slice out tensors using python's slice notation.

x = T[:, 0, None]
y = T[:, 1, None]

Internally, this delegates to Tensor.__getitem__, and is evaluated as part of the graph.