def test_infer_shape_gradI(self):
if theano.config.mode == "FAST_COMPILE":
raise SkipTest("Corr3dMM don't work in FAST_COMPILE")
def rand(*shape):
r = numpy.asarray(numpy.random.rand(*shape), dtype="float64")
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
gradI = corr3d.Corr3dMM_gradInputs
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [[1, 5, 6, 3, 3], [8, 2, 7, 3, 3], [1, 6, 9, 4, 4], [9, 6, 8, 5, 5], [9, 1, 6, 8, 8]]
bivec_vals = [[7, 5, 3, 1, 1], [4, 2, 5, 3, 3], [12, 6, 3, 2, 2], [5, 6, 1, 3, 3], [7, 1, 3, 4, 4]]
modes = ["valid", "full", "half", (1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2), 1]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode, subsample=subsample)(adtens, bdtens)
f = theano.function([adtens, bdtens], cdtens)
cdtens_val = f(adtens_val, bdtens_val)
# Corr3dMM_gradInputs
shape = (theano.shared(aivec_val[2]), theano.shared(aivec_val[3]), theano.shared(aivec_val[4]))
adtens_g = gradI(border_mode=mode, subsample=subsample)(bdtens, cdtens, shape=shape)
self._compile_and_check([bdtens, cdtens], [adtens_g], [bdtens_val, cdtens_val], gradI, warn=False)
def test_infer_shape_forward(self):
if theano.config.mode == "FAST_COMPILE":
raise SkipTest("Corr3dMM don't work in FAST_COMPILE")
def rand(*shape):
r = numpy.asarray(numpy.random.rand(*shape), dtype='float64')
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [[4, 5, 6, 3, 3], [6, 2, 8, 3, 3], [3, 6, 7, 5, 5],
[3, 6, 7, 5, 5], [5, 2, 4, 3, 3]]
bivec_vals = [[7, 5, 3, 2, 2], [4, 2, 5, 3, 3], [5, 6, 3, 2, 2],
[5, 6, 2, 3, 3], [6, 2, 4, 3, 3]]
modes = ['valid', 'full', 'half', (1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2), 1]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode, subsample=subsample)(adtens, bdtens)
self._compile_and_check([adtens, bdtens],
[cdtens],
[adtens_val, bdtens_val], corr3dMM,
warn=False)
def test_infer_shape_forward(self):
if theano.config.mode == "FAST_COMPILE":
raise SkipTest("Corr3dMM don't work in FAST_COMPILE")
def rand(*shape):
r = numpy.asarray(numpy.random.rand(*shape), dtype="float64")
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [[4, 5, 6, 3, 3], [6, 2, 8, 3, 3], [3, 6, 7, 5, 5], [3, 6, 7, 5, 5], [5, 2, 4, 3, 3]]
bivec_vals = [[7, 5, 3, 2, 2], [4, 2, 5, 3, 3], [5, 6, 3, 2, 2], [5, 6, 2, 3, 3], [6, 2, 4, 3, 3]]
modes = ["valid", "full", "half", (1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2), 1]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode, subsample=subsample)(adtens, bdtens)
self._compile_and_check([adtens, bdtens], [cdtens], [adtens_val, bdtens_val], corr3dMM, warn=False)
def test_infer_shape_gradI(self):
def rand(*shape):
r = np.asarray(np.random.rand(*shape), dtype="float64")
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
gradI = corr3d.Corr3dMMGradInputs
adtens = tt.dtensor5()
bdtens = tt.dtensor5()
aivec_vals = [
[1, 5, 6, 3, 3],
[8, 2, 7, 3, 3],
[1, 6, 9, 4, 4],
[9, 6, 8, 5, 5],
[9, 1, 6, 8, 8],
]
bivec_vals = [
[7, 5, 3, 1, 1],
[4, 2, 5, 3, 3],
[12, 6, 3, 2, 2],
[5, 6, 1, 3, 3],
[7, 1, 3, 4, 4],
]
modes = [
"valid", "full", "half", (1, 1, 1), (2, 1, 1), (1, 2, 1),
(1, 1, 2), 1
]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode,
subsample=subsample)(adtens, bdtens)
f = theano.function([adtens, bdtens], cdtens)
cdtens_val = f(adtens_val, bdtens_val)
# Corr3dMM_gradInputs
shape = (
theano.shared(aivec_val[2]),
theano.shared(aivec_val[3]),
theano.shared(aivec_val[4]),
)
adtens_g = gradI(border_mode=mode,
subsample=subsample)(bdtens,
cdtens,
shape=shape)
self._compile_and_check(
[bdtens, cdtens],
[adtens_g],
[bdtens_val, cdtens_val],
gradI,
warn=False,
)
def test_infer_shape_forward(self):
if theano.config.mode == "FAST_COMPILE":
pytest.skip("Corr3dMM don't work in FAST_COMPILE")
if not theano.config.cxx:
pytest.skip("Need cxx for this test")
def rand(*shape):
r = np.asarray(np.random.rand(*shape), dtype="float64")
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [
[4, 5, 6, 3, 3],
[6, 2, 8, 3, 3],
[3, 6, 7, 5, 5],
[3, 6, 7, 5, 5],
[5, 2, 4, 3, 3],
]
bivec_vals = [
[7, 5, 3, 2, 2],
[4, 2, 5, 3, 3],
[5, 6, 3, 2, 2],
[5, 6, 2, 3, 3],
[6, 2, 4, 3, 3],
]
modes = [
"valid", "full", "half", (1, 1, 1), (2, 1, 1), (1, 2, 1),
(1, 1, 2), 1
]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode,
subsample=subsample)(adtens, bdtens)
self._compile_and_check(
[adtens, bdtens],
[cdtens],
[adtens_val, bdtens_val],
corr3dMM,
warn=False,
)
def test_infer_shape_gradI(self):
if theano.config.mode == "FAST_COMPILE":
raise SkipTest("Corr3dMM don't work in FAST_COMPILE")
if not theano.config.cxx:
raise SkipTest("Need cxx for this test")
def rand(*shape):
r = np.asarray(np.random.rand(*shape), dtype='float64')
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
gradI = corr3d.Corr3dMM_gradInputs
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [[1, 5, 6, 3, 3], [8, 2, 7, 3, 3], [1, 6, 9, 4, 4],
[9, 6, 8, 5, 5], [9, 1, 6, 8, 8]]
bivec_vals = [[7, 5, 3, 1, 1], [4, 2, 5, 3, 3], [12, 6, 3, 2, 2],
[5, 6, 1, 3, 3], [7, 1, 3, 4, 4]]
modes = [
'valid', 'full', 'half', (1, 1, 1), (2, 1, 1), (1, 2, 1),
(1, 1, 2), 1
]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode,
subsample=subsample)(adtens, bdtens)
f = theano.function([adtens, bdtens], cdtens)
cdtens_val = f(adtens_val, bdtens_val)
# Corr3dMM_gradInputs
shape = (theano.shared(aivec_val[2]),
theano.shared(aivec_val[3]),
theano.shared(aivec_val[4]))
adtens_g = gradI(border_mode=mode,
subsample=subsample)(bdtens,
cdtens,
shape=shape)
self._compile_and_check([bdtens, cdtens], [adtens_g],
[bdtens_val, cdtens_val],
gradI,
warn=False)
def test_infer_shape_forward(self):
def rand(*shape):
r = np.asarray(np.random.rand(*shape), dtype="float64")
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
adtens = tt.dtensor5()
bdtens = tt.dtensor5()
aivec_vals = [
[4, 5, 6, 3, 3],
[6, 2, 8, 3, 3],
[3, 6, 7, 5, 5],
[3, 6, 7, 5, 5],
[5, 2, 4, 3, 3],
]
bivec_vals = [
[7, 5, 3, 2, 2],
[4, 2, 5, 3, 3],
[5, 6, 3, 2, 2],
[5, 6, 2, 3, 3],
[6, 2, 4, 3, 3],
]
modes = [
"valid", "full", "half", (1, 1, 1), (2, 1, 1), (1, 2, 1),
(1, 1, 2), 1
]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode,
subsample=subsample)(adtens, bdtens)
self._compile_and_check(
[adtens, bdtens],
[cdtens],
[adtens_val, bdtens_val],
corr3dMM,
warn=False,
)
def test_infer_shape_gradW(self):
if theano.config.mode == "FAST_COMPILE":
raise SkipTest("Corr3dMM don't work in FAST_COMPILE")
if not theano.config.cxx:
raise SkipTest("Need cxx for this test")
def rand(*shape):
r = np.asarray(np.random.rand(*shape), dtype='float64')
return r * 2 - 1
corr3dMM = corr3d.Corr3dMM
gradW = corr3d.Corr3dMM_gradWeights
adtens = T.dtensor5()
bdtens = T.dtensor5()
aivec_vals = [[1, 5, 6, 3, 3], [8, 2, 7, 3, 3], [1, 6, 9, 4, 4],
[9, 6, 8, 5, 5], [9, 1, 6, 8, 8]]
bivec_vals = [[7, 5, 3, 1, 1], [4, 2, 5, 3, 3], [12, 6, 3, 2, 2],
[5, 6, 1, 3, 3], [11, 1, 3, 3, 3]]
modes = ['valid', 'full', 'half', (1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2), 1]
subsamples = [(1, 1, 1), (2, 1, 1), (1, 2, 1), (1, 1, 2)]
for aivec_val, bivec_val in zip(aivec_vals, bivec_vals):
adtens_val = rand(*aivec_val)
bdtens_val = rand(*bivec_val)
for mode in modes:
for subsample in subsamples:
# Corr3dMM
cdtens = corr3dMM(border_mode=mode, subsample=subsample)(adtens, bdtens)
f = theano.function([adtens, bdtens], cdtens)
cdtens_val = f(adtens_val, bdtens_val)
# Corr3dMM_gradWeights
shape = (theano.shared(bivec_val[2]), theano.shared(bivec_val[3]),
theano.shared(bivec_val[4]))
bdtens_g = gradW(border_mode=mode,
subsample=subsample)(adtens, cdtens, shape=shape)
self._compile_and_check([adtens, cdtens],
[bdtens_g],
[adtens_val, cdtens_val], gradW,
warn=False)
