def test_conv_border_mode(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
with pytest.raises(Exception):
conv(X, W, (1, 1, 10), (2, 1, 3), border_mode=None)
def test_conv_stride(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
with pytest.raises(Exception):
conv(X, W, (1, 1, 10), (2, 1, 3), subsample=(2, ))
def test_conv_invalid_border_mode(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
with pytest.raises(Exception):
conv(X, W, (1, 1, 10), (2, 1, 3), border_mode=None)
def test_conv_stride(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
with pytest.raises(Exception):
conv(X, W, (1, 1, 10), (2, 1, 3), subsample=(2,))
def test_conv(impl, stride, filter_flip):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 10)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 6)))
conv_theano = conv(X, W, input.shape, kernel.shape, subsample=(stride,),
filter_flip=filter_flip).eval({X: input, W: kernel})
conv_np = conv1d(input, kernel, stride)
assert np.allclose(conv_theano, conv_np)
def test_conv_nones(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 12)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 3)))
conv_theano = conv(X, W, None, None).eval({
X: input, W: kernel
})
conv_np = conv1d(input, kernel)
assert np.allclose(conv_theano, conv_np)
def test_conv_pad(impl, pad):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 12)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 3)))
conv_theano = conv(X, W, input.shape, kernel.shape, border_mode=pad).eval({
X: input, W: kernel
})
pad = pad[0] if isinstance(pad, tuple) else pad
input = np.pad(input, [(0, 0), (0, 0), (pad, pad)], mode='constant')
conv_np = conv1d(input, kernel)
assert np.allclose(conv_theano, conv_np)
def test_conv_nones(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 12)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 3)))
conv_theano = conv(X, W, None, None).eval({X: input, W: kernel})
output = []
for b in input:
temp = []
for c in kernel:
temp.append(np.convolve(b[0, :], c[0, :], mode='valid'))
output.append(temp)
conv_np = np.array(output)
assert np.allclose(conv_theano, conv_np)
def test_conv_nones(impl):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 12)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 3)))
conv_theano = conv(X, W, None, None).eval({X: input, W: kernel})
output = []
for b in input:
temp = []
for c in kernel:
temp.append(np.convolve(b[0, :], c[0, :], mode="valid"))
output.append(temp)
conv_np = np.array(output)
assert np.allclose(conv_theano, conv_np)
def test_conv(impl, stride):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 10)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 6)))
conv_theano = conv(X, W, input.shape, kernel.shape, subsample=(stride,)).eval({X: input, W: kernel})
output = []
for b in input:
temp = []
for c in kernel:
temp.append(np.convolve(b[0, :], c[0, :], mode="valid"))
output.append(temp)
conv_np = np.array(output)[:, :, ::stride]
assert np.allclose(conv_theano, conv_np)
def test_conv(impl, stride):
import lasagne.theano_extensions.conv
conv = getattr(lasagne.theano_extensions.conv, impl)
X = T.tensor3()
W = T.tensor3()
input = lasagne.utils.floatX(np.ones((1, 1, 10)))
kernel = lasagne.utils.floatX(np.random.uniform(-1, 1, (2, 1, 6)))
conv_theano = conv(X, W, input.shape, kernel.shape,
subsample=(stride, )).eval({
X: input,
W: kernel
})
output = []
for b in input:
temp = []
for c in kernel:
temp.append(np.convolve(b[0, :], c[0, :], mode='valid'))
output.append(temp)
conv_np = np.array(output)[:, :, ::stride]
assert np.allclose(conv_theano, conv_np)
