def test_reduce():
sample_num = 200
def test_reduce_inner(numpy_reduce_func, nd_reduce_func, multi_axes):
for i in range(sample_num):
ndim = np.random.randint(1, 6)
shape = np.random.randint(1, 11, size=ndim)
dat = np.random.rand(*shape) - 0.5
keepdims = np.random.randint(0, 2)
if multi_axes:
axis_flags = np.random.randint(0, 2, size=ndim)
axes = []
for (axis, flag) in enumerate(axis_flags):
if flag:
axes.append(axis)
if 0 == len(axes):
axes = tuple(range(ndim))
else:
axes = tuple(axes)
else:
axes = np.random.randint(0, ndim)
numpy_ret = numpy_reduce_func(dat, axis=axes, keepdims=keepdims)
ndarray_ret = nd_reduce_func(mx.nd.array(dat),
axis=axes,
keepdims=keepdims)
if type(ndarray_ret) is mx.ndarray.NDArray:
ndarray_ret = ndarray_ret.asnumpy()
assert (ndarray_ret.shape == numpy_ret.shape) or \
(ndarray_ret.shape == (1,) and numpy_ret.shape == ()), "nd:%s, numpy:%s" \
%(ndarray_ret.shape, numpy_ret.shape)
err = np.square(ndarray_ret - numpy_ret).mean()
assert err < 1E-4
test_reduce_inner(
lambda data, axis, keepdims: np_reduce(data, axis, keepdims, np.sum),
mx.nd.sum, True)
test_reduce_inner(
lambda data, axis, keepdims: np_reduce(data, axis, keepdims, np.max),
mx.nd.max, True)
test_reduce_inner(
lambda data, axis, keepdims: np_reduce(data, axis, keepdims, np.min),
mx.nd.min, True)
# argmax and argmin are sensitive to the precision of the calculation (repro seed 1985162693).
# Force numpy to match mxnet's float32.
test_reduce_inner(
lambda data, axis, keepdims: np_reduce(np.float32(
data), axis, keepdims, np.argmax), mx.nd.argmax, False)
test_reduce_inner(
lambda data, axis, keepdims: np_reduce(np.float32(
data), axis, keepdims, np.argmin), mx.nd.argmin, False)
def test_reduce():
sample_num = 200
def test_reduce_inner(numpy_reduce_func, nd_reduce_func, multi_axes):
for i in range(sample_num):
ndim = np.random.randint(1, 6)
shape = np.random.randint(1, 11, size=ndim)
dat = np.random.rand(*shape) - 0.5
keepdims = np.random.randint(0, 2)
if multi_axes:
axis_flags = np.random.randint(0, 2, size=ndim)
axes = []
for (axis, flag) in enumerate(axis_flags):
if flag:
axes.append(axis)
if 0 == len(axes):
axes = tuple(range(ndim))
else:
axes = tuple(axes)
else:
axes = np.random.randint(0, ndim)
numpy_ret = numpy_reduce_func(dat, axis=axes, keepdims=keepdims)
ndarray_ret = nd_reduce_func(mx.nd.array(dat), axis=axes, keepdims=keepdims)
if type(ndarray_ret) is mx.ndarray.NDArray:
ndarray_ret = ndarray_ret.asnumpy()
assert (ndarray_ret.shape == numpy_ret.shape) or \
(ndarray_ret.shape == (1,) and numpy_ret.shape == ()), "nd:%s, numpy:%s" \
%(ndarray_ret.shape, numpy_ret.shape)
err = np.square(ndarray_ret - numpy_ret).mean()
assert err < 1E-4
test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.sum),
mx.nd.sum, True)
test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.max),
mx.nd.max, True)
test_reduce_inner(lambda data, axis, keepdims:np_reduce(data, axis, keepdims, np.min),
mx.nd.min, True)
# argmax and argmin are sensitive to the precision of the calculation (repro seed 1985162693).
# Force numpy to match mxnet's float32.
test_reduce_inner(lambda data, axis,
keepdims:np_reduce(np.float32(data), axis, keepdims, np.argmax),
mx.nd.argmax, False)
test_reduce_inner(lambda data, axis,
keepdims:np_reduce(np.float32(data), axis, keepdims, np.argmin),
mx.nd.argmin, False)