# Copyright 2019 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 mindspore.context as context import mindspore.nn as nn from mindspore import Tensor from mindspore.common.api import ms_function from mindspore.ops import operations as P context.set_context(device_target="Ascend") class Net(nn.Cell): def __init__(self, is_grad=False): super(Net, self).__init__() self.SparseSoftmaxCrossEntropyWithLogits = P.SparseSoftmaxCrossEntropyWithLogits(is_grad=is_grad) @ms_function def construct(self, features, labels): return self.SparseSoftmaxCrossEntropyWithLogits(features, labels) def np_sparse_softmax_cross_entropy_with_logits(labels_shape, logits_shape, logits_dtype): num_class = logits_shape[1] labels = np.random.randint(low=0, high=num_class - 1, size=labels_shape).astype(np.int32) logits = np.random.rand(*logits_shape).astype(logits_dtype) features = logits features_reshape = np.reshape(features, [-1, num_class]) labels_reshape = np.reshape(labels, [-1]) batch_dim = 0 class_dim = 1 batch_size = features_reshape.shape[batch_dim] e = np.exp(features_reshape - np.reshape(np.amax(features_reshape, axis=class_dim), [batch_size, 1])) probs = e / np.reshape(np.sum(e, axis=class_dim), [batch_size, 1]) labels_mat = np.zeros_like(probs).astype(probs.dtype) labels_mat[np.arange(batch_size), labels_reshape] = 1.0 bp = (probs - labels_mat) loss = -np.sum(labels_mat * np.log(probs + 1.0e-20), axis=1) bp_res = np.reshape(bp, features.shape) loss_res = np.reshape(loss, labels.shape) loss_res = np.sum(loss_res, axis=0) / loss_res.shape[0] return labels, logits, loss_res, bp_res def test_net(): '''Compare Numpy with MS type is float32''' labels_shape = (32,) logits_shape = [32, 1001] labels, logits, loss_np, _ = np_sparse_softmax_cross_entropy_with_logits(labels_shape, logits_shape, np.float32) expect = loss_np SparseSoftmaxCrossEntropyWithLogits = Net() loss_me = SparseSoftmaxCrossEntropyWithLogits(Tensor(logits), Tensor(labels)) # assert assert np.allclose(expect.flatten(), loss_me.asnumpy().flatten(), 0.01, 0.01) print(loss_me.asnumpy().flatten()) print("-------------------------") print(expect) test_net()