You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2018/05/14 18:22:44 UTC
[GitHub] szha closed pull request #10891: [MXNET-9845] Fix flakeyness of
test_operator.py:test_reduce.
szha closed pull request #10891: [MXNET-9845] Fix flakeyness of test_operator.py:test_reduce.
URL: https://github.com/apache/incubator-mxnet/pull/10891
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/tests/python/unittest/test_operator.py b/tests/python/unittest/test_operator.py
index 1db836b0918..c958c990f67 100644
--- a/tests/python/unittest/test_operator.py
+++ b/tests/python/unittest/test_operator.py
@@ -1843,6 +1843,10 @@ def test_reduce_inner(numpy_reduce_func, numpy_reduce_grad_func, mx_reduce_sym,
else:
b = mx_reduce_sym(a, axis=axes, keepdims=keepdims)
dat_npy = np.random.rand(*shape)
+ # Test with both negative and positive values (randomly). Avoid having both in the same
+ # test, which can be problematic for error checking due to near-zero values.
+ if np.random.rand() > 0.5:
+ dat_npy = -dat_npy
if nan_prob > 0:
dat_npy[np.random.rand(*shape) < nan_prob] = np.nan
sum_groundtruth = np.array(numpy_reduce_func(dat_npy, axis=axes, keepdims=keepdims))
@@ -1870,38 +1874,43 @@ def test_reduce_inner(numpy_reduce_func, numpy_reduce_grad_func, mx_reduce_sym,
test_none_axis = [True, False]
for test_none in test_none_axis:
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.sum),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape),
- mx.symbol.sum, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.mean),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape)/(data.size/outdata.size),
- mx.symbol.mean, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.prod),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape) * (outdata.reshape(keepdim_shape) / data),
- mx.symbol.prod, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.nansum),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- np.where(np.isnan(data), 0, outgrad.reshape(keepdim_shape)),
- mx.symbol.nansum, 0.3, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.nanprod),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- np.where(np.isnan(data), 0, outgrad.reshape(keepdim_shape) * (outdata.reshape(keepdim_shape) / data)),
- mx.symbol.nanprod, 0.3, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.max),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape) * (np.equal(data, outdata.reshape(keepdim_shape)).astype(np.float)),
- mx.symbol.max, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.min),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape) * (np.equal(data, outdata.reshape(keepdim_shape)).astype(np.float)),
- mx.symbol.min, test_none_axis=test_none)
- test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.linalg.norm),
- lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
- outgrad.reshape(keepdim_shape) * (data / outdata.reshape(keepdim_shape)),
- mx.symbol.norm, test_exclude=False, test_none_axis=test_none)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.sum),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape),
+ mx.symbol.sum, test_none_axis=test_none)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.mean),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape)/(data.size/outdata.size),
+ mx.symbol.mean, test_none_axis=test_none)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.prod),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape) * (outdata.reshape(keepdim_shape) / data),
+ mx.symbol.prod, test_none_axis=test_none)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.nansum),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ np.where(np.isnan(data), 0, outgrad.reshape(keepdim_shape)),
+ mx.symbol.nansum, 0.3, test_none_axis=test_none)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.nanprod),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ np.where(np.isnan(data), 0, outgrad.reshape(keepdim_shape) *
+ (outdata.reshape(keepdim_shape) / data)),
+ mx.symbol.nanprod, 0.3, test_none_axis=test_none)
+ # grad of max and min are sensitive to the precision of the calculation.
+ # Force numpy to match mxnet's float32.
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(np.float32(data), axis, keepdims, np.max),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape) *
+ (np.equal(np.float32(data), outdata.reshape(keepdim_shape))),
+ mx.symbol.max)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(np.float32(data), axis, keepdims, np.min),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape) *
+ (np.equal(np.float32(data), outdata.reshape(keepdim_shape))),
+ mx.symbol.min)
+ test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.linalg.norm),
+ lambda outgrad, data, outdata, axis, keepdims, keepdim_shape:
+ outgrad.reshape(keepdim_shape) * (data / outdata.reshape(keepdim_shape)),
+ mx.symbol.norm, test_exclude=False, test_none_axis=test_none)
@with_seed()
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services