• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1# Copyright 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# ==============================================================================
15"""
16Testing Vol op in DE
17"""
18import numpy as np
19import pytest
20
21import mindspore.dataset as ds
22import mindspore.dataset.audio.transforms as c_audio
23from mindspore import log as logger
24from mindspore.dataset.audio import utils
25
26
27def count_unequal_element(data_expected, data_me, rtol, atol):
28    """ Precision calculation func """
29    assert data_expected.shape == data_me.shape
30    total_count = len(data_expected.flatten())
31    error = np.abs(data_expected - data_me)
32    greater = np.greater(error, atol + np.abs(data_expected) * rtol)
33    loss_count = np.count_nonzero(greater)
34    assert (loss_count / total_count) < rtol, \
35        "\ndata_expected_std:{0}\ndata_me_error:{1}\nloss:{2}". \
36        format(data_expected[greater], data_me[greater], error[greater])
37
38
39def allclose_nparray(data_expected, data_me, rtol, atol, equal_nan=True):
40    """ Precision calculation formula  """
41    if np.any(np.isnan(data_expected)):
42        assert np.allclose(data_me, data_expected, rtol, atol, equal_nan=equal_nan)
43    elif not np.allclose(data_me, data_expected, rtol, atol, equal_nan=equal_nan):
44        count_unequal_element(data_expected, data_me, rtol, atol)
45
46
47def test_func_vol_eager():
48    """ mindspore eager mode normal testcase:vol op"""
49
50    logger.info("check vol op output")
51    ndarr_in = np.array([[0.3667, 0.5295, 0.2949, 0.4508, 0.6457, 0.3625, 0.4377, 0.3568],
52                         [0.6488, 0.6525, 0.5140, 0.6725, 0.9261, 0.0609, 0.3910, 0.4608],
53                         [0.0454, 0.0487, 0.6990, 0.1637, 0.5763, 0.1086, 0.5343, 0.4699],
54                         [0.9993, 0.0776, 0.3498, 0.0429, 0.1588, 0.3061, 0.1166, 0.3716],
55                         [0.7625, 0.2410, 0.8888, 0.5027, 0.0913, 0.2520, 0.5625, 0.9873]]).astype(np.float32)
56    # cal from benchmark
57    out_expect = np.array([[0.0733, 0.1059, 0.0590, 0.0902, 0.1291, 0.0725, 0.0875, 0.0714],
58                           [0.1298, 0.1305, 0.1028, 0.1345, 0.1852, 0.0122, 0.0782, 0.0922],
59                           [0.0091, 0.0097, 0.1398, 0.0327, 0.1153, 0.0217, 0.1069, 0.0940],
60                           [0.1999, 0.0155, 0.0700, 0.0086, 0.0318, 0.0612, 0.0233, 0.0743],
61                           [0.1525, 0.0482, 0.1778, 0.1005, 0.0183, 0.0504, 0.1125, 0.1975]])
62    op = c_audio.Vol(gain=0.2, gain_type=utils.GainType.AMPLITUDE)
63    out_mindspore = op(ndarr_in)
64    allclose_nparray(out_mindspore, out_expect, 0.0001, 0.0001)
65
66    ndarr_in = np.array([[[-0.5794799327850342, 0.19526369869709015],
67                          [-0.5935744047164917, 0.2948109209537506],
68                          [-0.42077431082725525, 0.04923877865076065]],
69                         [[0.5497273802757263, -0.22815021872520447],
70                          [-0.05891447141766548, -0.16206198930740356],
71                          [-1.4782767295837402, -1.3815662860870361]]]).astype(np.float32)
72    # cal from benchmark
73    out_expect = np.array([[[-0.5761537551879883, 0.1941428929567337],
74                            [-0.5901673436164856, 0.2931187152862549],
75                            [-0.41835910081863403, 0.04895615205168724]],
76                           [[0.5465719699859619, -0.22684065997600555],
77                            [-0.0585763081908226, -0.16113176941871643],
78                            [-1.0, -1.0]]])
79    op = c_audio.Vol(gain=-0.05, gain_type=utils.GainType.DB)
80    out_mindspore = op(ndarr_in)
81    allclose_nparray(out_mindspore, out_expect, 0.0001, 0.0001)
82
83    ndarr_in = np.array([[[0.09491927176713943, 0.11639882624149323, -0.1725238710641861, -0.18556903302669525],
84                          [-0.7140364646911621, 1.6223102807998657, 1.6710518598556519, 0.6019048094749451]],
85                         [[-0.8635917901992798, -0.31538113951683044, -0.2209240198135376, 1.3067045211791992],
86                          [-2.0922982692718506, 0.6822009682655334, 0.20066820085048676, 0.006392406765371561]]])
87    # cal from benchmark
88    out_expect = np.array([[[0.042449187487363815, 0.05205513536930084, -0.07715501636266708, -0.08298899233341217],
89                            [-0.31932681798934937, 0.7255191802978516, 0.7473170757293701, 0.2691799998283386]],
90                           [[-0.38620999455451965, -0.14104272425174713, -0.09880022704601288, 0.5843760371208191],
91                            [-0.935704231262207, 0.30508953332901, 0.0897415429353714, 0.0028587712440639734]]])
92    op = c_audio.Vol(gain=0.2, gain_type=utils.GainType.POWER)
93    out_mindspore = op(ndarr_in)
94    allclose_nparray(out_mindspore, out_expect, 0.0001, 0.0001)
95
96
97def test_func_vol_pipeline():
98    """ mindspore pipeline mode normal testcase:vol op"""
99
100    logger.info("test vol op with gain_type='power'")
101    data = np.array([[[0.7012, 0.2500, 0.0108],
102                      [0.3617, 0.6367, 0.6096]]]).astype(np.float32)
103    out_expect = np.array([[1.0000, 0.7906, 0.0342],
104                           [1.0000, 1.0000, 1.0000]])
105    data1 = ds.NumpySlicesDataset(data, column_names=["multi_dimensional_data"])
106    transforms = [c_audio.Vol(gain=10, gain_type=utils.GainType.POWER)]
107    data1 = data1.map(operations=transforms, input_columns=["multi_dimensional_data"])
108    for item in data1.create_dict_iterator(num_epochs=1, output_numpy=True):
109        out_put = item["multi_dimensional_data"]
110        allclose_nparray(out_put, out_expect, 0.0001, 0.0001)
111
112    logger.info("test vol op with gain_type='amplitude' and datatype='float64'")
113    data = np.array([[[0.9342139979247938, 0.613965955965896, 0.5356328030249583, 0.589909976354571],
114                      [0.7301220295167696, 0.31194499547960186, 0.3982210622160919, 0.20984374897512215],
115                      [0.18619300588033616, 0.9443723899839336, 0.7395507950492876, 0.4904588086175671]]])
116    data = data.astype(np.float64)
117    out_expect = np.array([[0.18684279918670654, 0.12279318571090699, 0.10712655782699586, 0.1179819941520691],
118                           [0.1460244059562683, 0.062388998270034794, 0.07964421510696412, 0.04196875095367432],
119                           [0.03723860085010529, 0.1888744831085205, 0.14791015386581421, 0.09809176325798036]])
120    data1 = ds.NumpySlicesDataset(data, column_names=["multi_dimensional_data"])
121    transforms = [c_audio.Vol(gain=0.2, gain_type=utils.GainType.AMPLITUDE)]
122    data1 = data1.map(operations=transforms, input_columns=["multi_dimensional_data"])
123    for item in data1.create_dict_iterator(num_epochs=1, output_numpy=True):
124        out_put = item["multi_dimensional_data"]
125        allclose_nparray(out_put, out_expect, 0.0001, 0.0001)
126
127    logger.info("test vol op with gain_type='db'")
128    data = np.array([[[0.1302, 0.5908, 0.1225, 0.7044],
129                      [0.6405, 0.6540, 0.9908, 0.8605],
130                      [0.7023, 0.0115, 0.8790, 0.5806]]]).astype(np.float32)
131    out_expect = np.array([[0.1096, 0.4971, 0.1031, 0.5927],
132                           [0.5389, 0.5503, 0.8336, 0.7240],
133                           [0.5909, 0.0097, 0.7396, 0.4885]])
134    data1 = ds.NumpySlicesDataset(data, column_names=["multi_dimensional_data"])
135    transforms = [c_audio.Vol(gain=-1.5, gain_type=utils.GainType.DB)]
136    data1 = data1.map(operations=transforms, input_columns=["multi_dimensional_data"])
137    for item in data1.create_dict_iterator(num_epochs=1, output_numpy=True):
138        out_put = item["multi_dimensional_data"]
139        allclose_nparray(out_put, out_expect, 0.0001, 0.0001)
140
141
142def test_vol_invalid_input():
143    def test_invalid_input(test_name, gain, gain_type, error, error_msg):
144        logger.info("Test Vol with invalid input: {0}".format(test_name))
145        with pytest.raises(error) as error_info:
146            c_audio.Vol(gain, gain_type)
147        assert error_msg in str(error_info.value)
148
149    test_invalid_input("invalid gain value when gain_type equals 'power'", -1.5, utils.GainType.POWER, ValueError,
150                       "Input gain is not within the required interval of (0, 16777216].")
151    test_invalid_input("invalid gain value when gain_type equals 'amplitude'", -1.5, utils.GainType.AMPLITUDE,
152                       ValueError, "Input gain is not within the required interval of [0, 16777216].")
153    test_invalid_input("invalid gain value when gain_type equals 'amplitude'", 1.5, "TEST", TypeError,
154                       "Argument gain_type with value TEST is not of type [<enum 'GainType'>], but got <class 'str'>.")
155
156
157if __name__ == "__main__":
158    test_func_vol_eager()
159    test_func_vol_pipeline()
160    test_vol_invalid_input()
161