@pblouw reported that his keyword spotting demo is broken in v0.3.0, and indeed our keyword spotting example is broken also. He tracked it down to this commit, which seems like it shouldn't even affect the keyword spotting network, but in fact it does.
The reason why that commit affects the keyword spotting network is that it translates all neuron->* connections in which
Investigating this issue is made more complicated by the fact that it is impossible to compare the two connection types because
One way we can still investigate this is in looking at neuron->* connections in which
Running that test with the commit it's introduced in (9042d86) reveals some hard to debug issues. Thanks to the new
I attempted to fix this in splitter in f2ae792. I was not able to fix it (I find the
It looks like the
Also, even if we fix this case, it doesn't fix the keyword spotting example. However, it might give us a bit of insight into why the switch from the
Rebased to #112 and reverted #85. I pushed the old branch to https://github.com/nengo/nengo-loihi/tree/fix-ks-saved. @tbekolay wrote a test and did some work there related to chip/host communication, but not directly related to the keyword demo, which we might want in the future.
tcstewar left a comment
This looks good to me as well. I will note that @pblouw seems to have fixed the keyword spotting demo when the connections are on-chip, so this PR might not be needed, but in the long run we're going to want a more robust way of controlling where the connection weights reside anyway.