tf.contrib.graph_editor.graph_replace()

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

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

Args:

• target_ts: a single tf.Tensor or an iterabble 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.