1# Copyright 2020-2021 Huawei Technologies Co., Ltd 2# 3# Licensed under the Apache License, Version 2.0 (the "License"); 4# you may not use this file except in compliance with the License. 5# You may obtain a copy of the License at 6# 7# http://www.apache.org/licenses/LICENSE-2.0 8# 9# Unless required by applicable law or agreed to in writing, software 10# distributed under the License is distributed on an "AS IS" BASIS, 11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12# See the License for the specific language governing permissions and 13# limitations under the License. 14# ============================================================================ 15import numpy as np 16import pytest 17 18import mindspore.nn as nn 19from mindspore import Tensor 20import mindspore.context as context 21from mindspore.ops import operations as P 22from mindspore.ops.operations import _inner_ops as inner 23 24class Net(nn.Cell): 25 def __init__(self, keep_prob): 26 super(Net, self).__init__() 27 self.drop = P.Dropout(keep_prob) 28 29 def construct(self, x_): 30 return self.drop(x_) 31 32 33@pytest.mark.level0 34@pytest.mark.platform_x86_gpu_training 35@pytest.mark.env_onecard 36def test_dropout(): 37 context.set_context(mode=context.PYNATIVE_MODE, device_target="GPU") 38 x_shape = [32, 16, 2, 5] 39 x = np.ones(x_shape).astype(np.float32) 40 keep_prob = 0.4 41 dropout = Net(keep_prob) 42 tx = Tensor(x) 43 output, mask = dropout(tx) 44 # check output 45 output_np = output.asnumpy() 46 elem_count = x.size 47 nonzero_count = np.count_nonzero(output_np) 48 assert (elem_count * (keep_prob - 0.1)) < nonzero_count < (elem_count * (keep_prob + 0.1)) 49 output_sum = np.sum(output_np) 50 x_sum = np.sum(x) 51 assert abs(output_sum - x_sum)/x_sum < 0.1 52 # check mask 53 mask_np = mask.asnumpy() 54 mask_sum = np.sum(mask_np) 55 assert np.count_nonzero(mask_np) == nonzero_count 56 assert abs(mask_sum - nonzero_count)/nonzero_count < 0.1 57 58 59class DropoutDynamic(nn.Cell): 60 def __init__(self, keep_prob): 61 super(DropoutDynamic, self).__init__() 62 self.test_dynamic = inner.GpuConvertToDynamicShape() 63 self.drop = P.Dropout(keep_prob) 64 65 def construct(self, x): 66 x = self.test_dynamic(x) 67 return self.drop(x) 68 69 70@pytest.mark.level0 71@pytest.mark.platform_x86_gpu_training 72@pytest.mark.env_onecard 73def test_dropout_dynamic(): 74 context.set_context(mode=context.GRAPH_MODE, device_target="GPU") 75 x_1 = np.ones([32, 16, 2, 5]).astype(np.float32) 76 x_2 = np.ones([32, 16, 2, 5, 6]).astype(np.float32) 77 keep_prob = 0.4 78 net = DropoutDynamic(keep_prob) 79 80 output_1, mask_1 = net(Tensor(x_1)) 81 elem_count_1 = x_1.size 82 nonzero_count_1 = np.count_nonzero(output_1.asnumpy()) 83 assert (elem_count_1 * (keep_prob - 0.1)) < nonzero_count_1 < (elem_count_1 * (keep_prob + 0.1)) 84 output_sum_1 = np.sum(output_1.asnumpy()) 85 x_sum_1 = np.sum(x_1) 86 assert abs(output_sum_1 - x_sum_1)/x_sum_1 < 0.1 87 mask_sum_1 = np.sum(mask_1.asnumpy()) 88 assert np.count_nonzero(mask_1.asnumpy()) == nonzero_count_1 89 assert abs(mask_sum_1 - nonzero_count_1)/nonzero_count_1 < 0.1 90 91 output_2, mask_2 = net(Tensor(x_2)) 92 elem_count_2 = x_2.size 93 nonzero_count_2 = np.count_nonzero(output_2.asnumpy()) 94 assert (elem_count_2 * (keep_prob - 0.1)) < nonzero_count_2 < (elem_count_2 * (keep_prob + 0.1)) 95 output_sum_2 = np.sum(output_2.asnumpy()) 96 x_sum_2 = np.sum(x_2) 97 assert abs(output_sum_2 - x_sum_2)/x_sum_2 < 0.1 98 mask_sum_2 = np.sum(mask_2.asnumpy()) 99 assert np.count_nonzero(mask_2.asnumpy()) == nonzero_count_2 100 assert abs(mask_sum_2 - nonzero_count_2)/nonzero_count_2 < 0.1 101