DeLesley Hutchins

I'm afraid I don't really understand your explanation of what you were trying to do. There are no tuples in your input -- just records and lists. The 'neighbors' field...

The original NEAT algorithm evolved small neural networks at the level of individual neurons. That obviously won't work with tensorflow; calling a python function to schedule a tensorflow operation to...

After traversing each tree, you need to call Weaver.add_output() on the root to mark it as one of the output tensors. All output tensors with the same shape will be...

Yes, you should use broadcast and zip. Broadcast does not copy the data; it's a python generator that expands a single element to a stream. On Mon, Apr 10, 2017...

To clarify slightly, using a metric will help if you want the output of the block to be a named output of the entire model; metric are automatically concatenated together....

You are correct; I wasn't thinking clearly. In fact, the input and output types of every block must have exact dimensions that are known in advance, so even if you...

If N is fixed and known, then you can also replace NGrams >> GetItem with SeqToTuple(T, N), where: def SeqToTuple(T, N): return (InputTransform(lambda x: tuple(x)) .set_input_type(td.SequenceType(T)) .set_output_type(td.Tuple(*([T] * N)))) which...

Your understanding is correct. This is an issue that we are very much aware of, but TensorFlow has some fundamental limitations that make it difficult. The problem essentially boils down...

I think the queue is a dead end for the reasons you mention. TensorFlow does have its own control-flow operators, so if you can confine the control-flow to within a...

sess.partial_run is the only reasonable way that TensorFlow currently provides for this sort of task. It's possible to trap the gradients yourself and feed them back in to another session.run,...