Skip to content

P
pytensor
提交 1af6f15d authored 作者: Pascal Lamblin's avatar Pascal Lamblin 提交者： GitHub
浏览文件
操作

Merge pull request #5097 from slefrancois/test_gpuarray_absconv

use floatX in gpuarray dnn tests
上级 baa3dd12 b6747150
隐藏空白字符变更
内嵌 并排
正在显示 包含 154 行增加 和 128 行删除
theano/gpuarray/tests/test_dnn.py
浏览文件 @ 1af6f15d
......@@ -24,6 +24,22 @@ from .rnn_support import Model, GRU, LSTM, WrapperLayer
from theano.configdefaults import SUPPORTED_DNN_CONV_ALGO_FWD
# If using float16, set CUDNN precision to float32
def set_precision(floatX):
if floatX == "float16":
precision = "float32"
else:
precision = theano.config.floatX
return precision
# If using float16, cast reference input to float32
def ref_cast(x):
if theano.config.floatX == 'float16':
x = T.cast(x, 'float32')
return x
def test_dnn_conv_desc_merge():
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
......@@ -50,18 +66,17 @@ def test_dnn_conv_merge():
raise SkipTest(dnn.dnn_available.msg)
img_shp = [2, 5, 6, 8]
kern_shp = [3, 5, 5, 6]
img = T.ftensor4('img')
kern = T.ftensor4('kern')
out = T.ftensor4('out')
desc = dnn.GpuDnnConvDesc(
border_mode='valid')(kern.shape)
img = T.tensor4('img')
kern = T.tensor4('kern')
out = T.tensor4('out')
desc = dnn.GpuDnnConvDesc(border_mode='valid')(kern.shape)
# Test forward op
o1 = dnn.dnn_conv(img, kern)
o2 = dnn.dnn_conv(img, kern)
f = theano.function([img, kern], [o1, o2], mode=mode_with_gpu)
d1, d2 = f(numpy.random.rand(*img_shp).astype('float32'),
numpy.random.rand(*kern_shp).astype('float32'))
d1, d2 = f(numpy.random.rand(*img_shp).astype(theano.config.floatX),
numpy.random.rand(*kern_shp).astype(theano.config.floatX))
topo = f.maker.fgraph.toposort()
assert len([n for n in topo if isinstance(n.op, dnn.GpuDnnConv)]) == 1
......@@ -89,9 +104,9 @@ def test_dnn_conv_inplace():
raise SkipTest(dnn.dnn_available.msg)
img_shp = [2, 5, 6, 8]
kern_shp = [3, 5, 5, 6]
img = T.ftensor4('img')
kern = T.ftensor4('kern')
out = T.ftensor4('out')
img = T.tensor4('img')
kern = T.tensor4('kern')
out = T.tensor4('out')
desc1 = dnn.GpuDnnConvDesc(border_mode='valid', conv_mode='conv')(
kern.shape)
desc2 = dnn.GpuDnnConvDesc(
......@@ -101,8 +116,8 @@ def test_dnn_conv_inplace():
o1 = dnn.dnn_conv(img, kern, conv_mode='conv')
o2 = dnn.dnn_conv(img, kern, conv_mode='cross')
f = theano.function([img, kern], [o1, o2], mode=mode_with_gpu)
d1, d2 = f(numpy.random.rand(*img_shp).astype('float32'),
numpy.random.rand(*kern_shp).astype('float32'))
d1, d2 = f(numpy.random.rand(*img_shp).astype(theano.config.floatX),
numpy.random.rand(*kern_shp).astype(theano.config.floatX))
topo = f.maker.fgraph.toposort()
convs = [n for n in topo if isinstance(n.op, dnn.GpuDnnConv)]
assert len(convs) == 2
......@@ -142,7 +157,7 @@ def test_pooling():
else:
modes = ('max', 'average_inc_pad', 'average_exc_pad')
x = T.ftensor4()
x = T.tensor4()
for mode, pad in product(modes,
((0, 0), (1, 0), (0, 1), (2, 3), (3, 2))):
if pad != (0, 0) and mode == 'average_exc_pad':
......@@ -180,7 +195,7 @@ def test_pooling():
(1, 3, 99, 99),
(32, 1, 147, 197),
]:
data = numpy.random.normal(0, 1, shp).astype("float32")
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX)
a = f_cpu(data).__array__()
b = f_gpu(data).__array__()
utt.assert_allclose(a, b)
......@@ -188,7 +203,7 @@ def test_pooling():
# Test the grad
for shp in [(1, 1, 2, 2),
(1, 1, 3, 3)]:
data = numpy.random.normal(0, 1, shp).astype("float32") * 10
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX) * 10
ws = 2
stride = 2
......@@ -226,7 +241,7 @@ def test_pooling():
def test_pooling_with_tensor_vars():
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
x = T.ftensor4()
x = T.tensor4()
ws = theano.shared(numpy.array([2, 2], dtype='int32'))
stride = theano.shared(numpy.array([1, 1], dtype='int32'))
pad = theano.shared(numpy.array([0, 0], dtype='int32'))
......@@ -242,7 +257,7 @@ def test_pooling_with_tensor_vars():
for shp in [(1, 1, 2, 2),
(1, 1, 3, 3)]:
data = numpy.random.normal(0, 1, shp).astype("float32") * 10
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX) * 10
theano.tests.unittest_tools.verify_grad(
fn, [data], mode=mode_with_gpu)
......@@ -266,7 +281,7 @@ def test_pooling_with_tensor_vars():
for shp in [(1, 10, 100, 100),
(1, 3, 99, 99),
(32, 1, 147, 197)]:
data = numpy.random.normal(0, 1, shp).astype("float32")
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX)
# Change the window size dynamically
ws.set_value(numpy.array([i, i]).astype('int32'))
......@@ -291,7 +306,7 @@ def test_pooling3d():
else:
modes = ('max', 'average_inc_pad', 'average_exc_pad')
x = T.ftensor5()
x = T.tensor5()
for mode, pad in product(modes,
((0, 0, 0), (1, 0, 0), (0, 1, 0), (0, 0, 1),
(2, 3, 2), (3, 2, 2), (2, 2, 3))):
......@@ -327,11 +342,11 @@ def test_pooling3d():
(1, 3, 99, 99, 29),
(2, 1, 147, 97, 37),
]:
data = numpy.random.normal(0, 1, shp).astype("float32")
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX)
a = f_cpu(data).__array__()
b = f_gpu(data).__array__()
utt.assert_allclose(a, b,
atol=numpy.finfo(numpy.float32).eps)
atol=numpy.finfo(theano.config.floatX).eps)
# Test the grad
for shp in [(1, 1, 2, 2, 2),
......@@ -341,7 +356,7 @@ def test_pooling3d():
(1, 1, 4, 3, 3),
(1, 1, 4, 4, 4),
(1, 1, 5, 5, 5)]:
data = numpy.random.normal(0, 1, shp).astype("float32") * 10
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX) * 10
ws = 2
stride = 2
......@@ -370,7 +385,7 @@ def test_pooling_opt():
raise SkipTest(dnn.dnn_available.msg)
# 2D pooling
x = T.fmatrix()
x = T.matrix()
f = theano.function(
[x],
......@@ -381,7 +396,7 @@ def test_pooling_opt():
assert any([isinstance(n.op, dnn.GpuDnnPool)
for n in f.maker.fgraph.toposort()])
f(numpy.zeros((10, 10), dtype='float32'))
f(numpy.zeros((10, 10), dtype=theano.config.floatX))
# gradient of 2D pooling
f = theano.function(
......@@ -394,7 +409,7 @@ def test_pooling_opt():
assert any([isinstance(n.op, dnn.GpuDnnPoolGrad)
for n in f.maker.fgraph.toposort()])
f(numpy.zeros((10, 10), dtype='float32'))
f(numpy.zeros((10, 10), dtype=theano.config.floatX))
# Test sum pooling
f = theano.function(
......@@ -405,11 +420,11 @@ def test_pooling_opt():
assert any([isinstance(n.op, dnn.GpuDnnPool)
for n in f.maker.fgraph.toposort()])
data = numpy.random.rand(10, 10).astype('float32')
data = numpy.random.rand(10, 10).astype(theano.config.floatX)
f(data)
# 3D pooling
x = T.ftensor3()
x = T.tensor3()
f = theano.function(
[x],
......@@ -420,7 +435,7 @@ def test_pooling_opt():
assert any([isinstance(n.op, dnn.GpuDnnPool)
for n in f.maker.fgraph.toposort()])
f(numpy.zeros((10, 10, 10), dtype='float32'))
f(numpy.zeros((10, 10, 10), dtype=theano.config.floatX))
# gradient of 3D pooling
f = theano.function(
......@@ -433,7 +448,7 @@ def test_pooling_opt():
assert any([isinstance(n.op, dnn.GpuDnnPoolGrad)
for n in f.maker.fgraph.toposort()])
f(numpy.zeros((10, 10, 10), dtype='float32'))
f(numpy.zeros((10, 10, 10), dtype=theano.config.floatX))
def test_pooling_opt_arbitrary_dimensions():
......@@ -454,7 +469,7 @@ def test_pooling_opt_arbitrary_dimensions():
# create input shape: non-pooling dimensions
# followed by 2 or 3 pooling dimensions
shp = tuple(range(2, 2 + n_non_pool_dims)) + tuple(range(5, 5 + len(ws)))
data = numpy.random.normal(0, 1, shp).astype('float32')
data = numpy.random.normal(0, 1, shp).astype(theano.config.floatX)
input = gpuarray_shared_constructor(data)
for mode in modes:
......@@ -491,7 +506,7 @@ def test_dnn_tag():
"""
Test that if cudnn isn't avail we crash and that if it is avail, we use it.
"""
x = T.ftensor4()
x = T.tensor4()
old = theano.config.on_opt_error
theano.config.on_opt_error = "raise"
......@@ -533,10 +548,10 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_softmax(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
t = T.ftensor4('t')
t = T.tensor4('t')
rand_tensor = numpy.asarray(
numpy.random.rand(5, 4, 3, 2),
dtype='float32'
dtype=theano.config.floatX
)
self._compile_and_check(
[t],
......@@ -564,19 +579,20 @@ class TestDnnInferShapes(utt.InferShapeTester):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img_val = numpy.asarray(img_val, dtype='float32')
kern_vals = numpy.asarray(kern_vals, dtype='float32')
img_val = numpy.asarray(img_val, dtype=theano.config.floatX)
kern_vals = numpy.asarray(kern_vals, dtype=theano.config.floatX)
for subsample in subsamples:
out_vals = numpy.zeros(
dnn.GpuDnnConv.get_out_shape(img_val.shape, kern_vals.shape,
border_mode=border_mode,
subsample=subsample),
dtype='float32')
dtype=theano.config.floatX)
desc = dnn.GpuDnnConvDesc(
border_mode=border_mode,
subsample=subsample,
conv_mode=conv_mode
conv_mode=conv_mode,
precision=set_precision(theano.config.floatX)
)(kerns.shape)
conv = dnn.GpuDnnConv(algo=algo)(img, kerns, out, desc)
self._compile_and_check(
......@@ -597,9 +613,9 @@ class TestDnnInferShapes(utt.InferShapeTester):
if algo == 'winograd' and dnn.version(raises=False) < 5000:
raise SkipTest(dnn.dnn_available.msg)
self._test_conv(T.ftensor4('img'),
T.ftensor4('kerns'),
T.ftensor4('out'),
self._test_conv(T.tensor4('img'),
T.tensor4('kerns'),
T.tensor4('out'),
numpy.random.rand(7, 2, 8, 4),
numpy.random.rand(8, 2, 4, 3),
border_mode,
......@@ -609,9 +625,9 @@ class TestDnnInferShapes(utt.InferShapeTester):
@parameterized.expand(product(border_modes, conv_modes), utt.custom_name_func)
def test_conv3d_none(self, border_mode, conv_mode):
self._test_conv(T.ftensor5('img'),
T.ftensor5('kerns'),
T.ftensor5('out'),
self._test_conv(T.tensor5('img'),
T.tensor5('kerns'),
T.tensor5('out'),
numpy.random.rand(10, 2, 6, 4, 11),
numpy.random.rand(8, 2, 4, 3, 1),
border_mode,
......@@ -625,11 +641,11 @@ class TestDnnInferShapes(utt.InferShapeTester):
img_val = numpy.asarray(
img_val,
dtype='float32'
dtype=theano.config.floatX
)
kern_vals = numpy.asarray(
kern_vals,
dtype='float32'
dtype=theano.config.floatX
)
temp_img = img.dimshuffle(1, 0, 2, 3)
......@@ -642,11 +658,12 @@ class TestDnnInferShapes(utt.InferShapeTester):
img_val.shape[2] - kern_vals.shape[2] + 1,
img_val.shape[3] - kern_vals.shape[3] + 1
)
out_vals = numpy.zeros(shape, dtype='float32')
out_vals = numpy.zeros(shape, dtype=theano.config.floatX)
desc = dnn.GpuDnnConvDesc(
border_mode=border_mode,
subsample=subsample,
conv_mode=conv_mode
conv_mode=conv_mode,
precision=set_precision(theano.config.floatX)
)(out.shape)
conv_grad_w = dnn.GpuDnnConvGradW()(
temp_img,
......@@ -663,9 +680,9 @@ class TestDnnInferShapes(utt.InferShapeTester):
@parameterized.expand(product(border_modes, conv_modes), utt.custom_name_func)
def test_conv_gradw(self, border_mode, conv_mode):
self._test_conv_gradw(T.ftensor4('img'),
T.ftensor4('kerns'),
T.ftensor4('out'),
self._test_conv_gradw(T.tensor4('img'),
T.tensor4('kerns'),
T.tensor4('out'),
numpy.random.rand(2, 5, 6, 8),
numpy.random.rand(2, 1, 5, 6),
border_mode,
......@@ -675,16 +692,16 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_conv_gradi(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor4('img')
kerns = T.ftensor4('kerns')
out = T.ftensor4('out')
img = T.tensor4('img')
kerns = T.tensor4('kerns')
out = T.tensor4('out')
kern_vals = numpy.asarray(
numpy.random.rand(13, 14, 15, 16),
dtype='float32'
dtype=theano.config.floatX
)
out_vals = numpy.asarray(
numpy.random.rand(3, 13, 5, 6),
dtype='float32'
dtype=theano.config.floatX
)
for params in product(
......@@ -697,11 +714,12 @@ class TestDnnInferShapes(utt.InferShapeTester):
out_vals.shape[2] + kern_vals.shape[2] - 1,
out_vals.shape[3] + kern_vals.shape[3] - 1
)
img_vals = numpy.zeros(shape, dtype='float32')
img_vals = numpy.zeros(shape, dtype=theano.config.floatX)
desc = dnn.GpuDnnConvDesc(
border_mode=params[0],
subsample=params[1],
conv_mode=params[2]
conv_mode=params[2],
precision=set_precision(theano.config.floatX)
)(kerns.shape)
conv_grad_i = dnn.GpuDnnConvGradI()(
kerns,
......@@ -719,10 +737,10 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_pool(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor4('img')
img = T.tensor4('img')
img_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5),
dtype='float32'
dtype=theano.config.floatX
)
# 'average_exc_pad' is disabled for versions < 4004
......@@ -746,10 +764,10 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_pool_3d(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor5('img')
img = T.tensor5('img')
img_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5, 6),
dtype='float32'
dtype=theano.config.floatX
)
# 'average_exc_pad' is disabled for versions < 4004
......@@ -773,20 +791,20 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_pool_grad(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor4('img')
img_grad = T.ftensor4('img_grad')
out = T.ftensor4('out')
img = T.tensor4('img')
img_grad = T.tensor4('img_grad')
out = T.tensor4('out')
img_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5),
dtype='float32'
dtype=theano.config.floatX
)
img_grad_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5),
dtype='float32'
dtype=theano.config.floatX
)
out_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5),
dtype='float32'
dtype=theano.config.floatX
)
for params in product(
......@@ -812,20 +830,20 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_pool_3d_grad(self):
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor5('img')
img_grad = T.ftensor5('img_grad')
out = T.ftensor5('out')
img = T.tensor5('img')
img_grad = T.tensor5('img_grad')
out = T.tensor5('out')
img_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5, 6),
dtype='float32'
dtype=theano.config.floatX
)
img_grad_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5, 6),
dtype='float32'
dtype=theano.config.floatX
)
out_val = numpy.asarray(
numpy.random.rand(2, 3, 4, 5, 6),
dtype='float32'
dtype=theano.config.floatX
)
for params in product(
......@@ -853,8 +871,8 @@ class TestDnnInferShapes(utt.InferShapeTester):
def test_dnn_conv_border_mode():
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor4()
kern = T.ftensor4()
img = T.tensor4()
kern = T.tensor4()
dnn.dnn_conv(img, kern, border_mode=1)
dnn.dnn_conv(img, kern, border_mode=(2, 3))
......@@ -866,9 +884,9 @@ def test_dnn_conv_border_mode():
def test_dnn_conv_alpha_output_merge():
if not dnn.dnn_available(test_ctx_name):
raise SkipTest(dnn.dnn_available.msg)
img = T.ftensor4()
kern = T.ftensor4()
out = T.ftensor4()
img = T.tensor4()
kern = T.tensor4()
out = T.tensor4()
b = 1
c = 4
......@@ -877,16 +895,16 @@ def test_dnn_conv_alpha_output_merge():
iw = 8
kh = 2
kw = 6
img_val = numpy.random.random((b, c, ih, iw)).astype('float32')
kern_val = numpy.random.random((f, c, kh, kw)).astype('float32')
img_val = numpy.random.random((b, c, ih, iw)).astype(theano.config.floatX)
kern_val = numpy.random.random((f, c, kh, kw)).astype(theano.config.floatX)
out_val = numpy.random.random((b, f, ih - kh + 1,
iw - kw + 1)).astype('float32')
iw - kw + 1)).astype(theano.config.floatX)
conv = dnn.dnn_conv(img, kern)
gw = theano.grad(conv.sum(), kern)
gi = theano.grad(conv.sum(), img)
lr = numpy.asarray(0.05, dtype='float32')
lr = numpy.asarray(0.05, dtype=theano.config.floatX)
fr = lr * (conv + out)
wr = kern + lr * gw
......@@ -936,31 +954,31 @@ def test_dnn_conv_grad():
iw = 8
kh = 2
kw = 2
img_val = numpy.random.random((b, c, ih, iw)).astype('float32')
kern_val = numpy.random.random((f, c, kh, kw)).astype('float32')
img_val = numpy.random.random((b, c, ih, iw)).astype(theano.config.floatX)
kern_val = numpy.random.random((f, c, kh, kw)).astype(theano.config.floatX)
out_val = numpy.random.random((b, f, ih - kw + 1,
iw - kw + 1)).astype('float32')
iw - kw + 1)).astype(theano.config.floatX)
def dconv(img, kern, out):
desc = dnn.GpuDnnConvDesc(border_mode='valid', subsample=(1, 1),
conv_mode='conv')(kern.shape)
conv_mode='conv', precision=set_precision(theano.config.floatX))(kern.shape)
return dnn.GpuDnnConv()(img, kern, out, desc, alpha=0.5, beta=0.75)
def dconvi(img, kern, out):
desc = dnn.GpuDnnConvDesc(border_mode='valid', subsample=(1, 1),
conv_mode='conv')(kern.shape)
conv_mode='conv', precision=set_precision(theano.config.floatX))(kern.shape)
return dnn.GpuDnnConvGradI()(kern, out, img, desc, alpha=-1.0,
beta=0.0)
def dconvw(img, kern, out):
desc = dnn.GpuDnnConvDesc(border_mode='valid', subsample=(1, 1),
conv_mode='conv')(kern.shape)
conv_mode='conv', precision=set_precision(theano.config.floatX))(kern.shape)
return dnn.GpuDnnConvGradW()(img, out, kern, desc, alpha=0.75,
beta=-1.0)
utt.verify_grad(dconv, [img_val, kern_val, out_val])
utt.verify_grad(dconvi, [img_val, kern_val, out_val])
utt.verify_grad(dconvw, [img_val, kern_val, out_val])
utt.verify_grad(dconv, [img_val, kern_val, out_val], eps=1e-3)
utt.verify_grad(dconvi, [img_val, kern_val, out_val], eps=1e-3)
utt.verify_grad(dconvw, [img_val, kern_val, out_val], eps=1e-3)
def get_conv3d_test_cases():
......@@ -1006,8 +1024,8 @@ def test_conv3d_fwd():
def run_conv3d_fwd(inputs_shape, filters_shape, subsample,
border_mode, conv_mode):
inputs_val = numpy.random.random(inputs_shape).astype('float32')
filters_val = numpy.random.random(filters_shape).astype('float32')
inputs_val = numpy.random.random(inputs_shape).astype(theano.config.floatX)
filters_val = numpy.random.random(filters_shape).astype(theano.config.floatX)
# Scale down the input values to prevent very large absolute errors
# due to float rounding
......@@ -1033,13 +1051,18 @@ def test_conv3d_fwd():
# Compile a theano function for the reference implementation
conv_ref = theano.tensor.nnet.corr3d.Corr3dMM(border_mode=border_mode,
subsample=subsample
)(inputs, flipped_filters)
)(ref_cast(inputs), flipped_filters)
f_ref = theano.function([], conv_ref, mode="FAST_RUN")
# Compare the results of the two implementations
res_ref = f_ref()
res = f()
utt.assert_allclose(res_ref, res)
# raise rtol to make the test pass with more seed.
rtol = None
# Raise tolerance for float16
if theano.config.floatX == 'float16':
rtol = 6e-2
utt.assert_allclose(res_ref, res, rtol=rtol)
test_cases = get_conv3d_test_cases()
for (i_shape, f_shape, subsample), border_mode, conv_mode in test_cases:
......@@ -1055,8 +1078,8 @@ def test_conv3d_bwd():
def run_conv3d_bwd(inputs_shape, filters_shape, subsample,
border_mode, conv_mode):
inputs_val = numpy.random.random(inputs_shape).astype('float32')
filters_val = numpy.random.random(filters_shape).astype('float32')
inputs_val = numpy.random.random(inputs_shape).astype(theano.config.floatX)
filters_val = numpy.random.random(filters_shape).astype(theano.config.floatX)
inputs = theano.shared(inputs_val)
filters = theano.shared(filters_val)
......@@ -1080,7 +1103,7 @@ def test_conv3d_bwd():
# Compile a theano function for the reference implementation
conv_ref = theano.tensor.nnet.corr3d.Corr3dMM(border_mode=border_mode,
subsample=subsample
)(inputs, flipped_filters)
)(ref_cast(inputs), flipped_filters)
(grad_i_ref,
grad_w_ref) = theano.tensor.grad(conv_ref.sum(),
[inputs, filters])
......@@ -1091,8 +1114,12 @@ def test_conv3d_bwd():
res = f()
# Needed for big size for some seed
# raise rtol to make the test pass with more seed.
utt.assert_allclose(res_ref[0], res[0], rtol=2e-5)
utt.assert_allclose(res_ref[1], res[1], rtol=2e-5)
rtol = None
# Raise tolerance for float16
if theano.config.floatX == 'float16':
rtol = 5e-2
utt.assert_allclose(res_ref[0], res[0], rtol=rtol)
utt.assert_allclose(res_ref[1], res[1], rtol=rtol)
test_cases = get_conv3d_test_cases()
for (i_shape, f_shape, subsample), border_mode, conv_mode in test_cases:
......@@ -1139,15 +1166,15 @@ class test_SoftMax(test_nnet.test_SoftMax):
def test_softmax_grad(self):
def cmp(n, m, f, f_gpu):
data = numpy.arange(n * m, dtype='float32').reshape(n, m)
data = numpy.arange(n * m, dtype=theano.config.floatX).reshape(n, m)
gdata = numpy.asarray(data)[:, :, None, None]
out = f(data)
gout = numpy.asarray(f_gpu(gdata))[:, :, 0, 0]
utt.assert_allclose(out, gout)
x = T.matrix('x', 'float32')
x_gpu = T.tensor4('x_gpu', 'float32')
x = T.matrix('x')
x_gpu = T.tensor4('x_gpu')
f_z = T.nnet.softmax_op
f_gpu = dnn.GpuDnnSoftmax(
'accurate',
......@@ -1158,7 +1185,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
dims = (2, 3, 4, 5)
gdata = numpy.arange(
numpy.product(dims),
dtype='float32'
dtype=theano.config.floatX
).reshape(dims)
T.verify_grad(f_gpu, [gdata], rng=numpy.random,
mode=mode_with_gpu)
......@@ -1180,14 +1207,14 @@ class test_SoftMax(test_nnet.test_SoftMax):
# Verify that the SoftmaxGrad -> Gpu[Dnn]SoftmaxGrad
# optimization is applied when cudnn is required
y = T.fvector('y')
y = T.vector('y')
f = theano.function(
[y],
T.grad(T.nnet.softmax(y).mean(), y),
mode=mode_with_gpu
)
sorted_f = f.maker.fgraph.toposort()
val = numpy.random.rand(5).astype('float32')
val = numpy.random.rand(5).astype(theano.config.floatX)
out_dnn = f(val)
assert(len([i
for i in sorted_f
......@@ -1206,7 +1233,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
# optimization is not applied when cudnn is excluded or not
# available
mode_wo_cudnn = mode_with_gpu.excluding("cudnn")
y = T.fvector('y')
y = T.vector('y')
f = theano.function(
[y],
T.grad(T.nnet.softmax(y).mean(), y),
......@@ -1230,7 +1257,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
# Verify that the SoftmaxGrad -> GpuDnnSoftmaxGrad do not
# crash with manual graph
y = T.fvector('y')
y = T.vector('y')
o = theano.tensor.nnet.SoftmaxGrad()(y, y * 2)
f = theano.function([y], o, mode=mode_with_gpu)
sorted_f = f.maker.fgraph.toposort()
......@@ -1253,7 +1280,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
if dnn.version(raises=False) < 3000:
raise SkipTest("Log-softmax is only in cudnn v3+")
x = T.ftensor4()
x = T.tensor4()
softmax_out = dnn.GpuDnnSoftmax('accurate', 'channel')(x)
log_out = T.log(T.as_tensor_variable(softmax_out))
......@@ -1277,7 +1304,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
(2, 3, 4, 66000)]
for inp_shape in input_shapes:
input_val = numpy.random.normal(0, 1, inp_shape).astype("float32")
input_val = numpy.random.normal(0, 1, inp_shape).astype(theano.config.floatX)
out = f(input_val)
expected_out = numpy.log(numpy.exp(input_val) /
......@@ -1296,7 +1323,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
# Compile a reference function, on the CPU, to be used to validate the
# results of the other function.
x = T.fmatrix()
x = T.matrix()
f_ref = theano.function([x], T.nnet.LogSoftmax()(x))
# Build the first graph and ensure that the optimization is applied
......@@ -1309,7 +1336,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
assert dnn_softmax_nodes[0].op.algo == "log"
# Compare the output of the function with the reference function
inp = numpy.random.normal(0, 1, (5, 6)).astype("float32")
inp = numpy.random.normal(0, 1, (5, 6)).astype(theano.config.floatX)
utt.assert_allclose(f(inp), f_ref(inp))
# Build the first graph and ensure that the optimization is applied
......@@ -1322,7 +1349,7 @@ class test_SoftMax(test_nnet.test_SoftMax):
assert dnn_softmax_nodes[0].op.algo == "log"
# Compare the output of the function with the reference function
inp = numpy.random.normal(0, 1, (5, 6)).astype("float32")
inp = numpy.random.normal(0, 1, (5, 6)).astype(theano.config.floatX)
utt.assert_allclose(f(inp), f_ref(inp))
......@@ -1334,7 +1361,7 @@ def test_dnn_batchnorm_train():
utt.seed_rng()
for mode in ('per-activation', 'spatial'):
for vartype in (T.ftensor5, T.ftensor4, T.ftensor3, T.fmatrix, T.fvector):
for vartype in (T.tensor5, T.tensor4, T.tensor3, T.matrix, T.vector):
x, scale, bias = (vartype(n) for n in ('x', 'scale', 'bias'))
ndim = x.ndim
eps = 5e-3 # some non-standard value to test if it's used
......@@ -1366,10 +1393,10 @@ def test_dnn_batchnorm_train():
data_shape = data_shape[:ndim]
param_shape = tuple(1 if d in axes else s
for d, s in enumerate(data_shape))
X = 4 + 3 * numpy.random.randn(*data_shape).astype('float32')
Dy = -1 + 2 * numpy.random.randn(*data_shape).astype('float32')
Scale = numpy.random.randn(*param_shape).astype('float32')
Bias = numpy.random.randn(*param_shape).astype('float32')
X = 4 + 3 * numpy.random.randn(*data_shape).astype(theano.config.floatX)
Dy = -1 + 2 * numpy.random.randn(*data_shape).astype(theano.config.floatX)
Scale = numpy.random.randn(*param_shape).astype(theano.config.floatX)
Bias = numpy.random.randn(*param_shape).astype(theano.config.floatX)
outputs = f(X, Scale, Bias, Dy)
# compare outputs
utt.assert_allclose(outputs[0], outputs[0 + 3]) # out
......@@ -1389,10 +1416,9 @@ def test_batchnorm_inference():
utt.seed_rng()
for mode in ('per-activation', 'spatial'):
for vartype in (T.ftensor5, T.ftensor4, T.ftensor3, T.fmatrix, T.fvector):
x, scale, bias, mean, var = (vartype(n) for n in ('x', 'scale',
'bias', 'mean',
'var'))
for vartype in (T.tensor5, T.tensor4, T.tensor3, T.matrix, T.vector):
x, scale, bias, mean, var = (vartype(n)
for n in ('x', 'scale', 'bias', 'mean', 'var'))
ndim = x.ndim
eps = 5e-3 # some non-standard value to test if it's used
......@@ -1420,12 +1446,12 @@ def test_batchnorm_inference():
data_shape = data_shape[:ndim]
param_shape = tuple(1 if d in axes else s
for d, s in enumerate(data_shape))
X = 4 + 3 * numpy.random.randn(*data_shape).astype('float32')
Dy = -1 + 2 * numpy.random.randn(*data_shape).astype('float32')
Scale = numpy.random.randn(*param_shape).astype('float32')
Bias = numpy.random.randn(*param_shape).astype('float32')
Mean = numpy.random.randn(*param_shape).astype('float32')
Var = numpy.random.rand(*param_shape).astype('float32')
X = 4 + 3 * numpy.random.randn(*data_shape).astype(theano.config.floatX)
Dy = -1 + 2 * numpy.random.randn(*data_shape).astype(theano.config.floatX)
Scale = numpy.random.randn(*param_shape).astype(theano.config.floatX)
Bias = numpy.random.randn(*param_shape).astype(theano.config.floatX)
Mean = numpy.random.randn(*param_shape).astype(theano.config.floatX)
Var = numpy.random.rand(*param_shape).astype(theano.config.floatX)
outputs = f(X, Scale, Bias, Mean, Var, Dy)
# compare outputs
utt.assert_allclose(outputs[0], outputs[1]) # out
......
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