tf.contrib.graph_editor.graph_replace

Create a new graph which compute the targets from the replaced Tensors.

tf.contrib.graph_editor.graph_replace(
    target_ts, replacement_ts, dst_scope='', src_scope='', reuse_dst_scope=False
)

Args
target_ts	a single tf.Tensor or an iterable of tf.Tensor.
replacement_ts	dictionary mapping from original tensors to replaced tensors
dst_scope	the destination scope.
src_scope	the source scope.
reuse_dst_scope	if True the dst_scope is re-used if it already exists. Otherwise, the scope is given a unique name based on the one given by appending an underscore followed by a digit (default).

Returns
A single tf.Tensor or a list of target tf.Tensor, depending on the type of the input argument target_ts. The returned tensors are recomputed using the tensors from replacement_ts.

Raises
ValueError	if the targets are not connected to replacement_ts.