Skip to content

P
pytensor
提交 966a535d authored 作者: affanv14's avatar affanv14 提交者： Mohammed Affan
浏览文件
操作

improve test code readability

上级 6888dbc8
隐藏空白字符变更
内嵌 并排
正在显示 包含 66 行增加 和 67 行删除
theano/tensor/nnet/tests/test_abstract_conv.py
浏览文件 @ 966a535d
...@@ -1721,58 +1721,58 @@ class Grouped_conv_noOptim(unittest.TestCase): ...@@ -1721,58 +1721,58 @@ class Grouped_conv_noOptim(unittest.TestCase):
raise SkipTest("CorrMM needs cxx and blas") raise SkipTest("CorrMM needs cxx and blas")
def test_fwd(self): def test_fwd(self):
img_sym = theano.tensor.tensor4('img')
kern_sym = theano.tensor.tensor4('kern')
for imshp, kshp, groups in zip(self.img_shape, self.kern_shape, self.num_groups): for imshp, kshp, groups in zip(self.img_shape, self.kern_shape, self.num_groups):
img = np.random.random(imshp).astype(theano.config.floatX) img = np.random.random(imshp).astype(theano.config.floatX)
kern = np.random.random(kshp).astype(theano.config.floatX) kern = np.random.random(kshp).astype(theano.config.floatX)
split_imgs = np.split(img, groups, axis=1) split_imgs = np.split(img, groups, axis=1)
split_kern = np.split(kern, groups, axis=0) split_kern = np.split(kern, groups, axis=0)
img_sym = theano.tensor.tensor4('img')
kern_sym = theano.tensor.tensor4('kern')
grouped_abstractconv_op = self.conv2d(border_mode=self.border_mode, grouped_conv_op = self.conv2d(border_mode=self.border_mode,
subsample=self.subsample, subsample=self.subsample,
filter_dilation=self.filter_dilation, filter_dilation=self.filter_dilation,
num_groups=groups) num_groups=groups)
if self.flip_filter: if self.flip_filter:
grouped_conv_output = grouped_abstractconv_op(img_sym, kern_sym[:, :, ::-1, ::-1]) grouped_conv_output = grouped_conv_op(img_sym, kern_sym[:, :, ::-1, ::-1])
else: else:
grouped_conv_output = grouped_abstractconv_op(img_sym, kern_sym) grouped_conv_output = grouped_conv_op(img_sym, kern_sym)
grouped_func = theano.function([img_sym, kern_sym], grouped_conv_output, mode=self.mode) grouped_func = theano.function([img_sym, kern_sym], grouped_conv_output, mode=self.mode)
assert any([isinstance(node.op, self.conv2d) assert any([isinstance(node.op, self.conv2d)
for node in grouped_func.maker.fgraph.toposort()]) for node in grouped_func.maker.fgraph.toposort()])
grouped_output = grouped_func(img, kern) grouped_output = grouped_func(img, kern)
normal_conv_op = conv2d_corr(img_sym, ref_conv_op = conv2d_corr(img_sym,
kern_sym, kern_sym,
border_mode=self.border_mode, border_mode=self.border_mode,
subsample=self.subsample, subsample=self.subsample,
filter_dilation=self.filter_dilation) filter_dilation=self.filter_dilation)
normal_func = theano.function([img_sym, kern_sym], normal_conv_op, ref_func = theano.function([img_sym, kern_sym], ref_conv_op,
mode=self.ref_mode) mode=self.ref_mode)
normal_concat_output = [normal_func(img_arr, kern_arr) ref_concat_output = [ref_func(img_arr, kern_arr)
for img_arr, kern_arr in zip(split_imgs, split_kern)] for img_arr, kern_arr in zip(split_imgs, split_kern)]
normal_concat_output = np.concatenate(normal_concat_output, axis=1) ref_concat_output = np.concatenate(ref_concat_output, axis=1)
utt.assert_allclose(grouped_output, normal_concat_output) utt.assert_allclose(grouped_output, ref_concat_output)
utt.verify_grad(grouped_abstractconv_op, utt.verify_grad(grouped_conv_op,
[img, kern], [img, kern],
mode=self.mode) mode=self.mode)
def test_gradweights(self): def test_gradweights(self):
img_sym = theano.tensor.tensor4('img')
top_sym = theano.tensor.tensor4('top')
for imshp, kshp, tshp, groups in zip(self.img_shape, self.kern_shape, self.top_shape, self.num_groups): for imshp, kshp, tshp, groups in zip(self.img_shape, self.kern_shape, self.top_shape, self.num_groups):
img = np.random.random(imshp).astype(theano.config.floatX) img = np.random.random(imshp).astype(theano.config.floatX)
top = np.random.random(tshp).astype(theano.config.floatX) top = np.random.random(tshp).astype(theano.config.floatX)
split_imgs = np.split(img, groups, axis=1) split_imgs = np.split(img, groups, axis=1)
split_top = np.split(top, groups, axis=1) split_top = np.split(top, groups, axis=1)
img_sym = theano.tensor.tensor4('img')
top_sym = theano.tensor.tensor4('top') grouped_convgrad_op = self.conv2d_gradw(border_mode=self.border_mode,
subsample=self.subsample,
grouped_abstractconvgrad_op = self.conv2d_gradw(border_mode=self.border_mode, filter_dilation=self.filter_dilation,
subsample=self.subsample, num_groups=groups)
filter_dilation=self.filter_dilation, grouped_conv_output = grouped_convgrad_op(img_sym, top_sym, kshp[-2:])
num_groups=groups)
grouped_conv_output = grouped_abstractconvgrad_op(img_sym, top_sym, kshp[-2:])
if self.flip_filter: if self.flip_filter:
grouped_conv_output = grouped_conv_output[:, :, ::-1, ::-1] grouped_conv_output = grouped_conv_output[:, :, ::-1, ::-1]
grouped_func = theano.function([img_sym, top_sym], grouped_conv_output, mode=self.mode) grouped_func = theano.function([img_sym, top_sym], grouped_conv_output, mode=self.mode)
...@@ -1780,67 +1780,66 @@ class Grouped_conv_noOptim(unittest.TestCase): ...@@ -1780,67 +1780,66 @@ class Grouped_conv_noOptim(unittest.TestCase):
for node in grouped_func.maker.fgraph.toposort()]) for node in grouped_func.maker.fgraph.toposort()])
grouped_output = grouped_func(img, top) grouped_output = grouped_func(img, top)
normal_conv_op = conv2d_corr_gw(img_sym, ref_conv_op = conv2d_corr_gw(img_sym,
top_sym, top_sym,
kshp, kshp,
border_mode=self.border_mode, border_mode=self.border_mode,
subsample=self.subsample, subsample=self.subsample,
filter_dilation=self.filter_dilation) filter_dilation=self.filter_dilation)
normal_func = theano.function([img_sym, top_sym], normal_conv_op, ref_func = theano.function([img_sym, top_sym], ref_conv_op,
mode=self.ref_mode) mode=self.ref_mode)
normal_concat_output = [normal_func(img_arr, top_arr) ref_concat_output = [ref_func(img_arr, top_arr)
for img_arr, top_arr in zip(split_imgs, split_top)] for img_arr, top_arr in zip(split_imgs, split_top)]
normal_concat_output = np.concatenate(normal_concat_output, axis=0) ref_concat_output = np.concatenate(ref_concat_output, axis=0)
utt.assert_allclose(grouped_output, normal_concat_output) utt.assert_allclose(grouped_output, ref_concat_output)
def abstract_conv_gradweight(inputs_val, output_val): def conv_gradweight(inputs_val, output_val):
return grouped_abstractconvgrad_op(inputs_val, output_val, kshp[-2:]) return grouped_convgrad_op(inputs_val, output_val, kshp[-2:])
utt.verify_grad(abstract_conv_gradweight, utt.verify_grad(conv_gradweight,
[img, top], [img, top],
mode=self.mode, eps=1) mode=self.mode, eps=1)
def test_gradinputs(self): def test_gradinputs(self):
kern_sym = theano.tensor.tensor4('kern')
top_sym = theano.tensor.tensor4('top')
for imshp, kshp, tshp, groups in zip(self.img_shape, self.kern_shape, self.top_shape, self.num_groups): for imshp, kshp, tshp, groups in zip(self.img_shape, self.kern_shape, self.top_shape, self.num_groups):
kern = np.random.random(kshp).astype(theano.config.floatX) kern = np.random.random(kshp).astype(theano.config.floatX)
top = np.random.random(tshp).astype(theano.config.floatX) top = np.random.random(tshp).astype(theano.config.floatX)
split_kerns = np.split(kern, groups, axis=0) split_kerns = np.split(kern, groups, axis=0)
split_top = np.split(top, groups, axis=1) split_top = np.split(top, groups, axis=1)
kern_sym = theano.tensor.tensor4('kern') grouped_convgrad_op = self.conv2d_gradi(border_mode=self.border_mode,
top_sym = theano.tensor.tensor4('top') subsample=self.subsample,
filter_dilation=self.filter_dilation,
grouped_abstractconvgrad_op = self.conv2d_gradi(border_mode=self.border_mode, num_groups=groups)
subsample=self.subsample,
filter_dilation=self.filter_dilation,
num_groups=groups)
if self.flip_filter: if self.flip_filter:
grouped_conv_output = grouped_abstractconvgrad_op(kern_sym[:, :, ::-1, ::-1], top_sym, imshp[-2:]) grouped_conv_output = grouped_convgrad_op(kern_sym[:, :, ::-1, ::-1], top_sym, imshp[-2:])
else: else:
grouped_conv_output = grouped_abstractconvgrad_op(kern_sym, top_sym, imshp[-2:]) grouped_conv_output = grouped_convgrad_op(kern_sym, top_sym, imshp[-2:])
grouped_func = theano.function([kern_sym, top_sym], grouped_conv_output, mode=self.mode) grouped_func = theano.function([kern_sym, top_sym], grouped_conv_output, mode=self.mode)
assert any([isinstance(node.op, self.conv2d_gradi) assert any([isinstance(node.op, self.conv2d_gradi)
for node in grouped_func.maker.fgraph.toposort()]) for node in grouped_func.maker.fgraph.toposort()])
grouped_output = grouped_func(kern, top) grouped_output = grouped_func(kern, top)
normal_conv_op = conv2d_corr_gi(kern_sym, ref_conv_op = conv2d_corr_gi(kern_sym,
top_sym, top_sym,
imshp, imshp,
border_mode=self.border_mode, border_mode=self.border_mode,
subsample=self.subsample, subsample=self.subsample,
filter_dilation=self.filter_dilation) filter_dilation=self.filter_dilation)
normal_func = theano.function([kern_sym, top_sym], normal_conv_op, ref_func = theano.function([kern_sym, top_sym], ref_conv_op,
mode=self.ref_mode) mode=self.ref_mode)
normal_concat_output = [normal_func(kern_arr, top_arr) ref_concat_output = [ref_func(kern_arr, top_arr)
for kern_arr, top_arr in zip(split_kerns, split_top)] for kern_arr, top_arr in zip(split_kerns, split_top)]
normal_concat_output = np.concatenate(normal_concat_output, axis=1) ref_concat_output = np.concatenate(ref_concat_output, axis=1)
utt.assert_allclose(grouped_output, normal_concat_output) utt.assert_allclose(grouped_output, ref_concat_output)
def abstract_conv_gradinputs(filters_val, output_val): def conv_gradinputs(filters_val, output_val):
return grouped_abstractconvgrad_op(filters_val, output_val, imshp[2:]) return grouped_convgrad_op(filters_val, output_val, imshp[2:])
utt.verify_grad(abstract_conv_gradinputs, utt.verify_grad(conv_gradinputs,
[kern, top], [kern, top],
mode=self.mode, eps=1) mode=self.mode, eps=1)
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论