denizyuret / Knet.jl

Koç University deep learning framework.
https://denizyuret.github.io/Knet.jl/latest
Other
1.43k stars 230 forks source link

test coverage is low #31

Open denizyuret opened 7 years ago

denizyuret commented 7 years ago

Add enough primitives to knetarray to pass the autograd tests converted to knetarrays. Add tests for cuda kernels in libknet8. Add code coverage service.

denizyuret commented 7 years ago

Emre, can you help with increasing the test coverage?

denizyuret commented 7 years ago

Started new unit testing structure, see test/cuda1.jl.

denizyuret commented 7 years ago

New tests in place as of [84fa45e 2017-02-12 distributions passes tests]. Knet/test has identically named files with Knet/src, each testing the functions in the other.

TODO:

Regarding the last point: my initial attempt at generating individual @tests with explicit function calls instead of the current for loops actually slowed test/unary.jl significantly. Need to investigate the speed issue more. Also need to understand how coverage measurement is done.

denizyuret commented 7 years ago

TODO: make each test run < 30 secs, currently linalg and update take > 60 secs.