#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Sign')
def ref_func_sign(x, alpha):
# returns -1 if x < 0, alpha if x==0, 1 if x > 0.
y = np.sign(x)
y[x == 0] = alpha
return y
def ref_grad_sign(x, dy, alpha):
# pass through gradient from output to input
return dy.flatten()
@pytest.mark.parametrize("ctx, func_name", ctxs)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Flip')
def ref_flip(x, axes):
if axes is None:
axes = [len(x.shape) - 1]
x = x.copy()
for axis in axes:
if axis == 0:
x = x[::-1]
elif axis == 1:
x = x[::, ::-1]
elif axis == 2:
x = x[::, ::, ::-1]
else:
raise
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('SoftmaxCrossEntropy')
def ref_softmax_cross_entropy(x, l, axis):
orig_x = x.copy()
x = x - x.max(axis, keepdims=True)
x = np.exp(x) / np.exp(x).sum(axis, keepdims=True)
x = np.rollaxis(x, axis, x.ndim).reshape(-1, x.shape[axis])
ll = np.rollaxis(l, axis, x.ndim).flatten()
y = - \
np.log(
np.maximum(x[np.arange(x.shape[0]), ll],
np.finfo(np.float32).tiny))
return y.reshape(l.shape)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import refs
from nbla_test_utils import list_context
ctxs = list_context('MaxPooling')
def ref_max_pooling(x, kernel, stride, ignore_border, pad):
# Only 2d
y = []
for xx in x.reshape((-1,) + x.shape[-3:]):
if xx.ndim == 2:
xx = xx[np.newaxis]
y += [refs.pooling_2d(xx, 'max', kernel, stride,
pad, ignore_border)[np.newaxis]]
y = np.vstack(y)
if x.ndim == 2:
y = np.squeeze(y, 1)
return y.reshape(x.shape[:-3] + y.shape[1:])
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('MinimumScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_minimum_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.minimum_scalar, np.minimum, inputs,
func_args=[val],
ctx=ctx, func_name=func_name)
@pytest.mark.parametrize("ctx, func_name", ctxs)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
from nnabla.testing import assert_allclose
ctxs = list_context('MeanSubtraction')
def ref_mean_subtraction(x, rmean, t, base_axis, batch_stat):
if batch_stat:
mean = x.mean(tuple(range(0, base_axis)))
rmean[...] = rmean + (mean - rmean) / (t + 1)
t += 1
return x - rmean
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape", [(2, 3, 4)])
@pytest.mark.parametrize("base_axis", [0, 1, 2])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_mean_subtraction_forward_backward(seed, inshape, base_axis, ctx,
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from six.moves import range
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('BatchNormalization')
def ref_batch_normalization(x, beta, gamma, rmean, rvar, axes, decay_rate,
eps, batch_stat, output_stat):
assert len(axes) == 1
reduc_axes = list(range(x.ndim))
del reduc_axes[axes[0]]
reduc_axes = tuple(reduc_axes)
m = x.size / x.shape[axes[0]]
if batch_stat:
mean = x.mean(reduc_axes, keepdims=True)
var = x.var(reduc_axes, keepdims=True)
rmean[...] = decay_rate * rmean + (1 - decay_rate) * mean
rvar[...] = decay_rate * rvar + (1 - decay_rate) * var * m / (m - 1)
else:
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('MulScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_mul_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng,
F.mul_scalar,
lambda x, y: x * y,
inputs,
func_args=[val],
ctx=ctx,
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from test_flip import ref_flip
from nbla_test_utils import list_context
ctxs = list_context('ImageAugmentation')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_image_augmentation_forward(seed, ctx, func_name):
rng = np.random.RandomState(seed)
inputs = [rng.randn(16, 3, 8, 8).astype(np.float32)]
i = nn.Variable(inputs[0].shape)
# NNabla forward
with nn.context_scope(ctx), nn.auto_forward():
o = F.image_augmentation(i)
assert o.d.shape == inputs[0].shape
shape = (3, 5, 8)
with nn.context_scope(ctx), nn.auto_forward():
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Constant')
def ref_constant(val, shape):
return np.ones(shape, dtype=np.float32) * val
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("value", [0., 1.5, -2., 1000.])
@pytest.mark.parametrize("shape", [[], [5], [2, 3, 4]])
def test_constant_forward(value, shape, ctx, func_name):
from nbla_test_utils import function_tester
rng = None
inputs = []
function_tester(rng, F.constant, ref_constant, inputs, func_args=[value, shape],
ctx=ctx, func_name=func_name, backward=[])
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Transpose')
def ref_transpose(x, axes):
return x.transpose(axes).copy()
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape, axes",
[((11, 13, 7), (2, 1, 0)),
((3, 7, 4, 5), (3, 0, 1, 2)),
((4, 2, 5, 2, 3), (3, 0, 1, 2, 4)),
((4, 2, 3, 2, 3, 3), (5, 3, 0, 1, 2, 4))])
def test_transpose_forward_backward(seed, inshape, axes, ctx, func_name):
from nbla_test_utils import function_tester
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('KLMultinomial')
def ref_kl_multinomial(p, q, base_axis):
kl = np.sum(p * (np.log(p + 1.0e-8) - np.log(q + 1.0e-8)),
axis=tuple(range(base_axis, p.ndim)))
return kl.reshape(kl.shape + (1,))
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("base_axis, shape", [(1, (3, 6)), (1, (5, 8, 7)), (2, (4, 7, 9))])
def test_kl_multinomial_forward_backward(seed, ctx, base_axis, shape, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
input0 = 1 + rng.rand(*(shape)).astype(np.float32)
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('MulScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_mul_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.mul_scalar, lambda x, y: x * y, inputs,
func_args=[val],
ctx=ctx, func_name=func_name)
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('CategoricalCrossEntropy')
def ref_categorical_cross_entropy(x, l, axis):
orig_x = x.copy()
x = np.rollaxis(x, axis, x.ndim).reshape(-1, x.shape[axis])
ll = np.rollaxis(l, axis, x.ndim).flatten()
y = - \
np.log(
np.maximum(x[np.arange(x.shape[0]), ll],
np.finfo(np.float32).tiny))
return y.reshape(l.shape)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Unpooling')
def ref_unpooling(x, kernel):
y = x
for ind, p in enumerate(kernel[::-1]):
y = y.repeat(p, axis=y.ndim - (ind + 1))
return y
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape", [(1, 2, 3), (2, 4, 6), (2, 2, 4, 6), (2, 2, 2, 4, 6)])
@pytest.mark.parametrize("kernel", [(1, 1), (2, 3), (2, 1, 2)])
def test_unpooling_forward_backward(seed, inshape, kernel, ctx, func_name):
from nbla_test_utils import function_tester
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Unlink')
def ref_unlink(x):
return x
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("shape", [(3, 6), (3, 7, 13)])
def test_unlink_forward_backward(seed, shape, ctx, func_name):
rng = np.random.RandomState(seed)
x0 = rng.randn(*(shape)).astype(np.float32)
ref = ref_unlink(x0)
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Log')
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_log_forward_backward(seed, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [
np.clip(np.abs(rng.randn(2, 3, 4).astype(np.float32)) * 1e10, 1e-6, 1e10)]
function_tester(rng, F.log, np.log, inputs, atol_f=1e-6,
atol_b=1e-2, ctx=ctx, func_name=func_name)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Exp')
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_exp_forward_backward(seed, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [np.clip(rng.randn(2, 3, 4).astype(np.float32) * 2.0, -2.0, 2.0)]
function_tester(rng,
F.exp,
np.exp,
inputs,
atol_f=1e-6,
atol_b=1e-2,
ctx=ctx,
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import division
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import refs
from nbla_test_utils import list_context
ctxs = list_context('Convolution')
def ref_convolution(x, w, b, base_axis, pad, stride, dilation, group):
y = []
for xx in x.reshape((-1,) + x.shape[base_axis:]):
y += [refs.convolution_2d(xx, w, b, pad, stride,
dilation, group)[np.newaxis]]
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape, kernel, outmaps, pad, stride, dilation",
[((2, 2, 10, 10), (3, 2), 4, (3, 0), (1, 2), (2, 1)),
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('BatchDet')
def ref_det(x):
y = np.zeros(x.shape[0], dtype=np.float32)
for i in range(x.shape[0]):
y[i] = np.linalg.det(x[i])
return y
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [314])
def test_batch_det_forward_backward(seed, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
# input must be batched square matrix
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import nnabla as nn
import nnabla.solvers as S
import numpy as np
from solver_test_utils import solver_tester, RefSolver
from nbla_test_utils import list_context
ctxs = list_context('Momentum')
class RefMomentum(RefSolver):
def __init__(self, lr, momentum):
self.lr = lr
self.momentum = momentum
self.v = {}
def _set_state_impl(self, key, param):
self.v[key] = np.zeros_like(param)
def _update_impl(self, key, p, g):
_update_momentum(p, g, self.v[key], self.lr, self.momentum)
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('RSubScalar')
def ref_r_sub_scalar(x, val):
return val - x
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [315])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_r_sub_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32)]
function_tester(rng,
F.r_sub_scalar,
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Softmax')
def ref_softmax(x, axis):
x = x - x.max(axis, keepdims=True)
x = np.exp(x) / np.exp(x).sum(axis, keepdims=True)
return x
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("axis", [0, 1, 2])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_softmax_forward_backward(seed, axis, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import nnabla.parametric_functions as PF
from nbla_test_utils import list_context
ctxs = list_context('GRU')
def gru(x, h, w, b, with_bias):
hidden_size = h.shape[1]
xh = F.concatenate(*(x, h), axis=1)
w0, w1, w2 = F.split(w, axis=0)
b0 = b1 = b2 = b3 = None
if with_bias:
b0, b1, b2, b3 = F.split(b, axis=0)
r_t = F.sigmoid(F.affine(xh, F.transpose(w0, (1, 0)), b0))
z_t = F.sigmoid(F.affine(xh, F.transpose(w1, (1, 0)), b1))
w2_0 = w2[:, :w2.shape[1] - hidden_size]
w2_1 = w2[:, w2.shape[1] - hidden_size:]
n_t = F.tanh(
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from test_flip import ref_flip
from nbla_test_utils import list_context
ctxs = list_context('RandomFlip')
"""
if hasattr(nn.extensions, 'cuda'):
ctxs += [(nn.extensions.cuda.context(), 'RandomFlipCuda')]
"""
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("axes", [None, (1,), (2,)])
def test_random_flip_forward_backward(seed, axes, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32)]
i = nn.Variable(inputs[0].shape, need_grad=True)
i.d = inputs[0]
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Reduce')
def ref_reduce(x, op):
return eval('np.%s' % op)(x)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("op", ['mean', 'sum'])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_reduce_forward_backward(seed, ctx, func_name, op):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.reduce, ref_reduce, inputs,
func_args=[op], ctx=ctx)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('ELU')
def ref_elu(x, alpha):
return np.maximum(x, 0) + alpha * (np.exp(np.minimum(x, 0)) - 1)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("alpha", [1.0, 0.5, 0.0])
@pytest.mark.parametrize("seed", [313])
def test_elu_forward_backward(seed, alpha, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.elu, ref_elu, inputs, func_args=[alpha],
ctx=ctx, func_name=func_name)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import nnabla as nn
import nnabla.solvers as S
import numpy as np
from solver_test_utils import solver_tester, RefSolver
from nbla_test_utils import list_context
ctxs = list_context('Sgd')
class RefSgd(RefSolver):
def __init__(self, lr):
self.lr = lr
def _set_state_impl(self, key, param):
pass
def _update_impl(self, key, p, g):
p[...] = p - self.lr * g
@pytest.mark.parametrize("ctx, solver_name", ctxs)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import refs
from nbla_test_utils import list_context
ctxs = list_context('SumPooling')
def ref_sum_pooling(x, kernel, stride, ignore_border, pad):
# Only 2d
y = []
for xx in x.reshape((-1,) + x.shape[-3:]):
if xx.ndim == 2:
xx = xx[np.newaxis]
y += [refs.pooling_2d(xx, 'sum', kernel, stride,
pad, ignore_border)[np.newaxis]]
y = np.vstack(y)
if x.ndim == 2:
y = np.squeeze(y, 1)
return y.reshape(x.shape[:-3] + y.shape[1:])
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Constant')
def ref_constant(val, shape):
return np.ones(shape, dtype=np.float32) * val
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("value", [0., 1.5, -2., 1000.])
@pytest.mark.parametrize("shape", [[], [5], [2, 3, 4]])
def test_constant_forward(value, shape, ctx, func_name):
from nbla_test_utils import function_tester
rng = None
inputs = []
function_tester(rng,
F.constant,
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('RDivScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [314])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_r_div_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester, cap_ignore_region
rng = np.random.RandomState(seed)
inputs = [
cap_ignore_region(
rng.randn(2, 3, 4).astype(np.float32) * 3, (-0.5, 0.5))]
function_tester(rng, F.r_div_scalar, lambda x, y: y / x, inputs,
func_args=[val], dstep=1e-4, atol_b=1e-1,
ctx=ctx, func_name=func_name)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('CELU')
def ref_celu(x, alpha, axis):
elu1 = np.maximum(0., x) + alpha * (np.exp(np.minimum(0, x)) - 1)
elu2 = np.maximum(0., -x) + alpha * (np.exp(np.minimum(0, -x)) - 1)
return np.concatenate([elu1, elu2], axis=axis)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("alpha", [1.0, 0.5, 0.0])
@pytest.mark.parametrize("axis", [0, 1, 2, -1, -2, -3])
@pytest.mark.parametrize("seed", [313])
def test_celu_forward_backward(seed, alpha, axis, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('VATNoise')
def ref_vat_noise(r, base_axis, eps):
shape = r.shape
r = r.reshape(shape[0:base_axis] + (np.prod(shape[base_axis:]), ))
r = r / np.sqrt(np.sum(r**2, axis=base_axis)
).reshape(shape[0:base_axis] + (1,))
r = r.reshape(shape).astype(np.float32)
return eps * r
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("base_axis, shape", [(1, (3, 6)), (1, (5, 8)), (2, (3, 7, 13))])
@pytest.mark.parametrize("eps", [10, 1.4])
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Identity')
def ref_identity(x):
return x
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_identity_forward_backward(seed, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32)]
function_tester(rng, F.identity, ref_identity, inputs,
ctx=ctx, func_name=func_name, atol_b=1e-3)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('MeanSubtraction')
def ref_mean_subtraction(x, rmean, t, base_axis, batch_stat):
if batch_stat:
mean = x.mean(tuple(range(0, base_axis)))
rmean[...] = rmean + (mean - rmean) / (t + 1)
t += 1
return x - rmean
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape", [(2, 3, 4)])
@pytest.mark.parametrize("base_axis", [0, 1, 2])
@pytest.mark.parametrize("ctx, func_name", ctxs)
def test_mean_subtraction_forward_backward(seed, inshape, base_axis, ctx, func_name):
ref_func,
inputs,
func_args=[axis],
func_kwargs=dict(keepdims=keepdims),
ctx=ctx,
func_name=func_name,
# The backward test on macOS doesn't pass with this tolerance.
# Does Eigen library used in CPU computation backend produce
# the different results on different platforms?
# atol_b=3e-3,
atol_b=6e-3)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape, axis", [((2, 1, 64, 64), (2, 3))])
@pytest.mark.parametrize("ctx, func_name", list_context('Sum'))
def test_large_sum_reduction(seed, inshape, axis, ctx, func_name):
if not func_name.endswith('Cuda'):
# This configuration is only to test the CUDA implementation branch
# where reduction_size / outer_size >= 2048, so we skip this test for
# CPU and CuDNN and also do not need to run for both keepdims.
pytest.skip('skip CUDA specific implementation test for other target')
rng = np.random.RandomState(seed)
inputs = [rng.randn(*inshape).astype(np.float32)]
function_tester(rng,
F.sum,
np.sum,
inputs,
ctx=ctx,
func_name=func_name,
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import division
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import refs
from nbla_test_utils import list_context
ctxs = list_context('BinaryConnectConvolution')
def binarize(x):
y = np.sign(x)
y[y == 0] = -1.0
return y
def ref_convolution(x, w, b, base_axis, pad, stride, dilation, group):
y = []
for xx in x.reshape((-1,) + x.shape[base_axis:]):
y += [refs.convolution_2d(xx, w, b, pad, stride,
dilation, group)[np.newaxis]]
y = np.vstack(y)
return y.reshape(x.shape[:base_axis] + y.shape[1:])
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Concatenate')
def ref_concatenate(*inputs, **params):
axis = params.pop('axis', len(inputs[0].shape) - 1)
return np.concatenate(inputs, axis=axis)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("axis", [0, 1, 2])
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize('different_size', [False, True])
@pytest.mark.parametrize('num_inputs', [2, 3])
def test_concatenate_forward_backward(seed, axis, different_size, num_inputs,
ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Reshape')
def ref_reshape(x, shape):
return x.reshape(shape).copy()
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape, outshape", [((1, 1, 6), (1, 2, 3)), ((2, 3), (1, 6))])
def test_reshape_forward_backward(seed, inshape, outshape, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
# Input
inputs = [rng.randn(*inshape).astype(np.float32)]
function_tester(rng, F.reshape, ref_reshape, inputs, func_args=[
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('ACos')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_acos_forward_backward(seed, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [np.clip(rng.randn(2, 3, 4).astype(np.float32), -0.9, 0.9)]
function_tester(rng,
F.acos,
np.arccos,
inputs,
ctx=ctx,
func_name=func_name,
atol_f=1e-3,
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Embed')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("shape_x", [(10, ), (2, 8), (2, 3, 4), (2, 2, 3, 4)])
@pytest.mark.parametrize("shape_w", [(5, 3), (4, 3, 4), (6, 2, 2, 3)])
def test_embed_forward_backward(seed, shape_x, shape_w, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
n_class = shape_w[0]
x = rng.randint(0, n_class - 1, shape_x).astype(np.int32)
w = rng.randn(*shape_w).astype(np.float32)
inputs = [x, w]
function_tester(rng,
F.embed,
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
from nnabla.testing import assert_allclose
ctxs = list_context('TopNError')
def ref_top_n_error(x, l, axis, n):
orig_x = x.copy()
x = np.rollaxis(x, axis, x.ndim).reshape(-1, x.shape[axis])
ll = np.rollaxis(l, axis, x.ndim).flatten()
y = []
for x_, ll_ in zip(x, ll):
threshold = x_[ll_]
count = 0
for x__ in x_:
if x__ >= threshold:
count += 1
y.append(1 if count > n else 0)
return np.array(y).reshape(l.shape)
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from __future__ import division
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
import refs
from nbla_test_utils import list_context
ctxs = list_context('BinaryConnectConvolution')
def binarize(x, quantize_zero_to):
y = np.sign(x)
y[y == 0] = quantize_zero_to
return y
def ref_convolution(x, w, b, base_axis, pad, stride, dilation, group,
quantize_zero_to):
y = []
for xx in x.reshape((-1, ) + x.shape[base_axis:]):
y += [
refs.convolution_2d(xx, w, b, pad, stride, dilation,
group)[np.newaxis]
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import nnabla.solvers as S
import numpy as np
from solver_test_utils import solver_tester, RefSolver
from nbla_test_utils import list_context
ctxs = list_context('Nesterov')
class RefNesterov(RefSolver):
def __init__(self, lr, momentum):
self.lr = lr
self.momentum = momentum
self.v = {}
def _set_state_impl(self, key, param):
self.v[key] = np.zeros_like(param)
def _update_impl(self, key, p, g):
_update_nesterov(p, g, self.v[key], self.lr, self.momentum)
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
This is still left unlike other *2 operation (sub2, mul2, ...) because
it has cudnn implementation.
"""
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Minimum2')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_minimum2_double_backward(seed, ctx, func_name):
from nbla_test_utils import backward_function_tester
rng = np.random.RandomState(seed)
inputs = [
rng.randn(2, 3).astype(np.float32),
rng.randn(2, 3).astype(np.float32)
]
backward_function_tester(rng,
F.minimum2,
None,
inputs=inputs,
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""
This is still left unlike other *2 operation (sub2, mul2, ...) because
it has cudnn implementation.
"""
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Maximum2')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_maximum2_double_backward(seed, ctx, func_name):
from nbla_test_utils import backward_function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3).astype(np.float32),
rng.randn(2, 3).astype(np.float32)]
backward_function_tester(rng, F.maximum2,
inputs=inputs,
func_args=[], func_kwargs={},
atol_accum=1e-3,
dstep=1e-3,
ctx=ctx)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs_rand = list_context('Rand')
ctxs_randint = list_context('Randint')
ctxs_randn = list_context('Randn')
ctxs_rand_binomial = list_context('RandBinomial')
ctxs_rand_beta = list_context('RandBeta')
ctxs_rand_gamma = list_context('RandGamma')
@pytest.mark.parametrize("ctx, func_name", ctxs_rand)
@pytest.mark.parametrize("low, high", [(0, 1), (-2.5, 100), (0.1, 0.11)])
@pytest.mark.parametrize("shape", [[], [5], [100, 100]])
@pytest.mark.parametrize("seed", [-1, 313])
def test_rand_forward(seed, ctx, func_name, low, high, shape):
with nn.context_scope(ctx):
o = F.rand(low, high, shape, seed=seed)
assert o.shape == tuple(shape)
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Reshape')
def ref_reshape(x, shape, inplace):
return x.reshape(shape).copy()
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape, outshape", [((1, 1, 6), (1, 2, 3)),
((2, 3), (1, 6)),
((2, 4), (-1, 2, 2))])
def test_reshape_forward_backward(seed, inshape, outshape, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
# Input
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Pow2')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inplace", [False, True])
def test_pow2_double_backward(inplace, seed, ctx, func_name):
from nbla_test_utils import backward_function_tester
rng = np.random.RandomState(seed)
inputs = [
(rng.randint(5, size=(2, 3)).astype(np.float32) + 1.0) * 0.75,
rng.randn(2, 3).astype(np.float32),
]
backward_function_tester(rng,
F.pow2,
inputs=inputs,
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Sinh')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_sinh_forward_backward(seed, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 1]
function_tester(rng, F.sinh, np.sinh, inputs, ctx=ctx, func_name=func_name,
atol_f=1e-3, atol_b=1e-2)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_sinh_double_backward(seed, ctx, func_name):
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('NormNormalization')
def ref_norm_normalization(x, p, axis, eps=1e-12):
if p is None:
p = 2.
# calculate norm
y = x
y = np.abs(y)
y = np.power(y, p)
y = np.sum(y, axis, keepdims=True) + eps
y = np.power(y, 1. / p)
# normalization
y = x / y
return y
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('RPowScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [314])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_r_pow_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.r_pow_scalar, lambda x, y: y ** x, inputs,
func_args=[val], atol_b=1e-2,
ctx=ctx, func_name=func_name)
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import nnabla as nn
import nnabla.solvers as S
import numpy as np
from solver_test_utils import solver_tester, RefSolver
from nbla_test_utils import list_context
ctxs = list_context('Adam')
class RefAdam(RefSolver):
def __init__(self, alpha, beta1, beta2, eps):
self.alpha = alpha
self.beta1 = beta1
self.beta2 = beta2
self.eps = eps
self.m = {}
self.v = {}
self.t = {}
def _set_state_impl(self, key, param):
self.m[key] = np.zeros_like(param)
self.v[key] = np.zeros_like(param)
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('AddScalar')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("val", [0.5, 1, 2])
def test_add_scalar_forward_backward(seed, val, ctx, func_name):
from nbla_test_utils import function_tester
rng = np.random.RandomState(seed)
inputs = [rng.randn(2, 3, 4).astype(np.float32) * 2]
function_tester(rng, F.add_scalar, lambda x, y: x + y, inputs,
func_args=[val],
ctx=ctx, func_name=func_name)
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context, function_tester
ctxs = list_context('TopKGrad')
def ref_top_k_grad(k, abs, base_axis, grad):
outer_dim = np.prod(grad.shape[:base_axis], dtype=int)
inner_dim = np.prod(grad.shape[base_axis:], dtype=int)
gg = grad.reshape(outer_dim, inner_dim).copy()
ix = np.argsort(np.abs(gg) if abs else gg)[:, -k:]
dx = np.zeros((outer_dim, inner_dim))
for idx, row in enumerate(ix):
dx[idx, row] = gg[idx, row]
dx = dx.squeeze(axis=0) if base_axis == 0 else dx
return dx.reshape(grad.shape)
def ref_top_k_grad_fw(x, k, abs, base_axis):
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('EpsilonInsensitiveLoss')
def ref_epsilon_insensitive_loss_forward(x0, x1, epsilon):
y = np.zeros_like(x0)
abs_diff = np.abs(x0 - x1)
idx = np.where(abs_diff > epsilon)
y[idx] = abs_diff[idx] - epsilon
return y
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("epsilon", [0.001, 1])
def test_epsilon_insensitive_loss_forward_backward(seed, ctx, func_name, epsilon):
from nbla_test_utils import function_tester
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Abs')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
def test_abs_forward_backward(seed, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [
cap_ignore_region(
rng.randn(2, 3, 4).astype(np.float32) * 2,
(-1e-3, 1e-3))]
function_tester(rng, F.abs, np.abs, inputs,
ctx=ctx, func_name=func_name)
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import numpy as np
from scipy.ndimage.interpolation import shift as scipy_shift
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Shift')
def ref_shift(x, shifts, border_mode):
if shifts is None:
shifts = (0,) * len(x.shape)
return scipy_shift(x, shifts, mode=border_mode)
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("inshape", [(2, 3, 4)])
@pytest.mark.parametrize("shifts", [None, (0, 0, 2,), (0, -2, 1), (1, 0, -1)])
@pytest.mark.parametrize("border_mode", ["nearest", "reflect"])
def test_shift_forward_backward(seed, inshape, shifts, border_mode, ctx, func_name):
from nbla_test_utils import function_tester
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from six.moves import range
import pytest
import numpy as np
import nnabla as nn
import nnabla.functions as F
from nbla_test_utils import list_context
ctxs = list_context('Dropout')
@pytest.mark.parametrize("ctx, func_name", ctxs)
@pytest.mark.parametrize("seed", [313])
@pytest.mark.parametrize("p", [p / 10. for p in range(1, 9)])
def test_dropout_forward_backward(p, seed, ctx, func_name):
from nbla_test_utils import cap_ignore_region, function_tester
rng = np.random.RandomState(seed)
inputs = [
cap_ignore_region(
rng.randn(2, 3, 4).astype(np.float32) * 2,
(-1e-3, 1e-3))] # Ensure there is no zero.
i = nn.Variable(inputs[0].shape, need_grad=True)
i.d = inputs[0]
# NNabla forward
