Add unit test framework for operators' conversion #29
Conversation
There was a problem hiding this comment.
Already implemented the unit test for conv2d, pool2d, mum_op etc, but not for batch_norm_op due to its special unit test.
| @@ -24,6 +24,6 @@ unittest(){ | |||
| trap 'abort' 0 | |||
| set -e | |||
|
|
|||
| unittest . | |||
| # unittest . | |||
There was a problem hiding this comment.
The unit test in travis-ci is disabled because it requires the installation of caffe2, as the backend of ONNX.
There was a problem hiding this comment.
Yes, seems caffe2 backend is necessary, let me try to add the dependencies in the CI.
There was a problem hiding this comment.
@pkuyym the installation of caffe2 always failed in local machine, and seems that it is very sensitive to the dependencies, e.g. version of cmake, protobuf etc. You can find if there is a robust way, so that we can integrate it in CI. Thanks!
| 'use_mkldnn': self.use_mkldnn | ||
| } | ||
|
|
||
| output = np.zeros((1, 1, 1, 1)) |
There was a problem hiding this comment.
output is a placeholder, and can be any shape and any value.
|
|
||
| self.inputs = {'Input': input, 'Filter': filter} | ||
|
|
||
| self.persistable = ['Filter'] |
There was a problem hiding this comment.
This list of persistable vars, usually indicating the parameters
|
Really enjoying this PR... comments coming soon.. |
|
There's a lot of good things going on here. Before I proceed, have you considered adding |
|
@varunarora of course. Done adding |
| for name, np_value in self.inputs[var_name]: | ||
| tensor = core.LoDTensor() | ||
| if isinstance(np_value, tuple): | ||
| tensor.set(np_value[0], place) |
There was a problem hiding this comment.
It would be nice to make new function for this code that repeats twice doing the .sets
There was a problem hiding this comment.
It is true. I will fix it in next pull request.
|
@varunarora I saw the comments you added and learn a lot from them. This is a good practice. Many thanks! |
* add base test case * add run script * update * add pytest configruation
Fix #28