add st test_pynative_control_flow.py

pull/8886/head
wanghua 4 years ago
parent 85a020575a
commit 4acdcad5d6

@ -0,0 +1,121 @@
# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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 numpy as np
import pytest
import mindspore.context as context
import mindspore.nn as nn
from mindspore import Tensor
import mindspore.common.dtype as mstype
import mindspore.ops as P
from mindspore.common import ParameterTuple
import torch
import torch.nn as nn_pt
context.set_context(mode=context.PYNATIVE_MODE, device_target="Ascend")
class GradofAllInputsAndParams(nn.Cell):
def __init__(self, net, sens=False):
super().__init__()
self.grad = P.GradOperation(get_all=True, get_by_list=True, sens_param=sens)
self.net = net
self.params = ParameterTuple(self.net.trainable_params())
def construct(self, *x):
out = self.grad(self.net, self.params)(*x)
return out
@pytest.mark.level0
@pytest.mark.platform_arm_ascend_training
@pytest.mark.platform_x86_ascend_training
@pytest.mark.env_onecard
def test_pynative_diff_shape_with_while_in_construct():
class WhileNetMs(nn.Cell):
def __init__(self):
super().__init__()
self.conv = nn.Conv2d(1, 1, 3, weight_init='ones', pad_mode='pad')
def construct(self, x, flag):
while flag:
if flag > 1:
x = self.conv(x)
else:
x = x + 1
flag = flag - 1
return x
class WhileNetPt(nn_pt.Module):
def __init__(self):
super().__init__()
self.conv = nn_pt.Conv2d(in_channels=1, out_channels=1, kernel_size=(3, 3),
stride=1, padding=0, bias=False)
self.weight = nn_pt.Parameter(torch.from_numpy(np.ones([1, 1, 3, 3]).astype(np.float32)))
self.conv.register_parameter('weight', self.weight)
def forward(self, x, flag):
while flag:
if flag > 1:
x = self.conv(x)
else:
x = x + 1
flag = flag - 1
return x
net = WhileNetMs()
input_ms = Tensor(np.random.rand(1, 1, 224, 224).astype(np.float32))
flag = 2
out = net(input_ms, flag)
backnet = GradofAllInputsAndParams(net)
backout = backnet(input_ms, Tensor(flag, mstype.int32))
comparenet = WhileNetPt()
torch_input = torch.from_numpy(input_ms.asnumpy())
torch_input.requires_grad = True
torch_flag = torch.from_numpy(np.array(flag))
torch_flag.requires_grad = False
out_good = comparenet(torch_input, torch_flag)
grad = torch.from_numpy(np.ones_like(out_good.detach().numpy()).astype(np.float32))
out_good.backward(gradient=grad)
assert np.allclose(out_good.detach().numpy(), out.asnumpy(), 0.01, 0.01)
assert np.allclose(torch_input.grad.numpy(), backout[0][0].asnumpy(), 0.01, 0.01)
assert np.allclose(comparenet.weight.grad.numpy(), backout[1][0].asnumpy(), 0.01, 0.01)
flag = 3
out = net(input_ms, flag)
backout = backnet(input_ms, Tensor(flag, mstype.int32))
torch_flag = torch.from_numpy(np.array(flag))
torch_flag.requires_grad = False
comparenet.zero_grad()
torch_input.grad.zero_()
out_good = comparenet(torch_input, torch_flag)
grad = torch.from_numpy(np.ones_like(out_good.detach().numpy()).astype(np.float32))
out_good.backward(gradient=grad)
assert np.allclose(out_good.detach().numpy(), out.asnumpy(), 0.01, 0.01)
assert np.allclose(torch_input.grad.numpy(), backout[0][0].asnumpy(), 0.01, 0.01)
assert np.allclose(comparenet.weight.grad.numpy(), backout[1][0].asnumpy(), 0.01, 0.01)
input_ms = Tensor(np.random.rand(1, 1, 112, 112).astype(np.float32))
flag = 4
backout = backnet(input_ms, Tensor(flag, mstype.int32))
torch_input = torch.from_numpy(input_ms.asnumpy())
torch_input.requires_grad = True
torch_flag = torch.from_numpy(np.array(flag))
torch_flag.requires_grad = False
comparenet.zero_grad()
out_good = comparenet(torch_input, torch_flag)
grad = torch.from_numpy(np.ones_like(out_good.detach().numpy()).astype(np.float32))
out_good.backward(gradient=grad)
assert np.allclose(torch_input.grad.numpy(), backout[0][0].asnumpy(), 0.01, 0.01)
assert np.allclose(comparenet.weight.grad.numpy(), backout[1][0].asnumpy(), 0.01, 0.01)
Loading…
Cancel
Save