Repository navigation
Ensure things happen in consistent order - #151
Conversation
tbekolay
left a comment
tbekolay
left a comment
There was a problem hiding this comment.
Looking good! Made some inline comments, and this needs a changelog.
It would also be nice if there was a test, but I have to admit I am struggling to think of a test that isn't trivial ... do you think this will end up getting tested in #70?
| """For receiving host->chip messages""" | ||
|
|
||
| def __init__(self, dimensions, size_out): | ||
| def __init__(self, dimensions, size_out, **kwargs): |
There was a problem hiding this comment.
I think **kwargs is confusing here; for me, I don't really remember what other parameters there are to Node. If I look it up and see size_in, I'll get a TypeError if I try to specify it. I think we should bite the bullet and expose label and seed (or just label if seeds are irrelevant here).
| class ChipReceiveNeurons(ChipReceiveNode): | ||
| """Passes spikes directly (no on-off neuron encoding)""" | ||
| def __init__(self, dimensions, neuron_type=None): | ||
| def __init__(self, dimensions, neuron_type=None, **kwargs): |
There was a problem hiding this comment.
Here also, explicitly include label and (perhaps) seed.
|
From the review: we should add a test that splits a model multiple times and ensures that their order in the split networks is the same. |
|
I will write this test. |
c7e42b6 to
d83169c
Compare
|
Ok, I've added a test. I've tested that it can distinguish between before this PR and after it. |
- Splitter preserves object order from original network. - Host/chip receivers are always called in same order. Also added optional labels to ChipReceiveNodes and their CxSpikeInputs.
a381cf7 to
cd3ba5a
Compare
drasmuss
left a comment
drasmuss
left a comment
There was a problem hiding this comment.
Looks good to me, just waiting for tests to run then will merge.
3 participants
This makes sure that seeds are always set the same as well. Right now, setting a seed on a network does not actually guarantee it will be the same from one run to the next, because the splitter dictionaries change the order that everything gets built in in a non-deterministic way.