1# Copyright 2020 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# ============================================================================ 15""" test lstm """ 16import pytest 17 18import mindspore.context as context 19from mindspore import nn 20from ..ut_filter import run_on_gpu 21from ....ops_common import convert 22 23 24class LstmTestNet(nn.Cell): 25 """ LstmTestNet definition """ 26 27 def __init__(self, input_size, hidden_size, num_layers, has_bias, batch_first, bidirectional): 28 super(LstmTestNet, self).__init__() 29 self.lstm = nn.LSTM(input_size=input_size, 30 hidden_size=hidden_size, 31 num_layers=num_layers, 32 has_bias=has_bias, 33 batch_first=batch_first, 34 bidirectional=bidirectional, 35 dropout=0.0) 36 37 def construct(self, inp, h0, c0): 38 return self.lstm(inp, (h0, c0)) 39 40 41test_case_cell_ops = [ 42 ('lstm1_with_bias', { 43 'cell': LstmTestNet(10, 12, 2, has_bias=True, batch_first=False, bidirectional=False), 44 'input_shape': [[5, 3, 10], [2, 3, 12], [2, 3, 12]], 45 'output_shape': [[5, 3, 12], [2, 3, 12], [2, 3, 12]]}), 46 ('lstm2_without_bias', { 47 'cell': LstmTestNet(10, 12, 2, has_bias=False, batch_first=False, bidirectional=False), 48 'input_shape': [[5, 3, 10], [2, 3, 12], [2, 3, 12]], 49 'output_shape': [[5, 3, 12], [2, 3, 12], [2, 3, 12]]}), 50 ('lstm3_with_bias_bidirectional', { 51 'cell': LstmTestNet(10, 12, 2, has_bias=True, batch_first=False, bidirectional=True), 52 'input_shape': [[5, 3, 10], [4, 3, 12], [4, 3, 12]], 53 'output_shape': [[5, 3, 24], [4, 3, 12], [4, 3, 12]]}), 54 ('lstm4_without_bias_bidirectional', { 55 'cell': LstmTestNet(10, 12, 2, has_bias=False, batch_first=False, bidirectional=True), 56 'input_shape': [[5, 3, 10], [4, 3, 12], [4, 3, 12]], 57 'output_shape': [[5, 3, 24], [4, 3, 12], [4, 3, 12]]}), 58 ('lstm5_with_bias_batch_first', { 59 'cell': LstmTestNet(10, 12, 2, has_bias=True, batch_first=True, bidirectional=False), 60 'input_shape': [[3, 5, 10], [2, 3, 12], [2, 3, 12]], 61 'output_shape': [[3, 5, 12], [2, 3, 12], [2, 3, 12]]}), 62 ('lstm6_without_bias_batch_first', { 63 'cell': LstmTestNet(10, 12, 2, has_bias=False, batch_first=True, bidirectional=False), 64 'input_shape': [[3, 5, 10], [2, 3, 12], [2, 3, 12]], 65 'output_shape': [[3, 5, 12], [2, 3, 12], [2, 3, 12]]}), 66 ('lstm7_with_bias_bidirectional_batch_first', { 67 'cell': LstmTestNet(10, 12, 2, has_bias=True, batch_first=True, bidirectional=True), 68 'input_shape': [[3, 5, 10], [4, 3, 12], [4, 3, 12]], 69 'output_shape': [[3, 5, 24], [4, 3, 12], [4, 3, 12]]}), 70 ('lstm8_without_bias_bidirectional_batch_first', { 71 'cell': LstmTestNet(10, 12, 2, has_bias=False, batch_first=True, bidirectional=True), 72 'input_shape': [[3, 5, 10], [4, 3, 12], [4, 3, 12]], 73 'output_shape': [[3, 5, 24], [4, 3, 12], [4, 3, 12]]}), 74] 75 76 77# use -k to select certain testcast 78# pytest tests/python/ops/test_lstm.py::test_compile -k lstm_with_bias 79 80@pytest.mark.parametrize('args', test_case_cell_ops, ids=lambda x: x[0]) 81def test_compile(args): 82 config = args[1] 83 shapes = config['input_shape'] 84 net = config['cell'] 85 net.set_train() 86 inputs = [convert(shp) for shp in shapes] 87 out = net(*inputs) 88 print(f"out: {out}") 89 90 91@run_on_gpu 92@pytest.mark.parametrize('args', test_case_cell_ops, ids=lambda x: x[0]) 93def test_execute(args): 94 """ test_execute """ 95 config = args[1] 96 shapes = config['input_shape'] 97 net = config['cell'] 98 net.set_train() 99 inputs = [convert(shp) for shp in shapes] 100 context.set_context(mode=context.GRAPH_MODE, device_target="GPU") 101 # pylint: disable=unused-variable 102 ret, (hn, cn) = net(*inputs) 103 print(f'result: {shapes[0]} --> {ret.asnumpy().shape}, expected: {config["output_shape"][0]}') 104