Closed hunse closed 5 years ago
A number of tests have started failing for weird reasons. test_learning.test_multiple_pes
fails sporadically; hopefully with #151 it will be more predictable.
There's also failing for the static and docs tests, because of weird problems with the setup for CI. Not sure if it's because something changed with packages in pip, or how the Travis servers are initialized, or what.
Also, the codecov is bad because the new tests are importorskip('nengo_dl')
, and it appears they're not getting run. But nengo_dl
should be installed for the tests, I thought. I think it might have to do with the version (I might be using some newer nengo_dl
features).
All LGTM now!
Other than those two coverage points, the changes look good to me.
This adds nengo_dl TensorFlow builders for the LoihiLIF and LoihiSpikingRectifiedLinear neuron types. This allows nengo_dl networks to train with tuning curves like those of the chip neurons, allowing for more accurate conversion of ANNs for the chip.
Based off #133