Coverage for /scratch/builds/bob/bob.med.tb/miniconda/conda-bld/bob.med.tb_1637571489937/_test_env_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placehold_placeho/lib/python3.8/site-packages/bob/med/tb/test/test_config.py: 62%

1#!/usr/bin/env python 

2# coding=utf-8 

3 

4import torch 

5import pytest 

6from torch.utils.data import ConcatDataset 

7import numpy as np 

8import contextlib 

9from bob.extension import rc 

10 

11from ..configs.datasets import get_samples_weights, get_positive_weights 

12 

13from . import mock_dataset 

14 

15# Download test data and get their location if needed 

16montgomery_datadir = mock_dataset() 

17 

18# we only iterate over the first N elements at most - dataset loading has 

19# already been checked on the individual datset tests. Here, we are only 

20# testing for the extra tools wrapping the dataset 

21N = 10 

22 

23 

24@contextlib.contextmanager 

25def rc_context(**new_config): 

26 old_rc = rc.copy() 

27 rc.update(new_config) 

28 try: 

29 yield 

30 finally: 

31 rc.clear() 

32 rc.update(old_rc) 

33 

34 

35@pytest.mark.skip_if_rc_var_not_set("bob.med.tb.montgomery.datadir") 

36def test_montgomery(): 

37 

38 def _check_subset(samples, size): 

39 assert len(samples) == size 

40 for s in samples[:N]: 

41 assert len(s) == 3 

42 assert isinstance(s[0], str) #key 

43 assert s[1].shape == (1, 512, 512) #planes, height, width 

44 assert s[1].dtype == torch.float32 

45 assert isinstance(s[2], int) #label 

46 assert s[1].max() <= 1.0 

47 assert s[1].min() >= 0.0 

48 

49 from ..configs.datasets.montgomery.default import dataset 

50 

51 assert len(dataset) == 4 

52 _check_subset(dataset["__train__"], 110) 

53 _check_subset(dataset["__valid__"], 110) 

54 _check_subset(dataset["train"], 110) 

55 _check_subset(dataset["test"], 28) 

56 

57def test_get_samples_weights(): 

58 

59 # Temporarily modify Montgomery datadir 

60 new_value = {"bob.med.tb.montgomery.datadir": montgomery_datadir} 

61 with rc_context(**new_value): 

62 

63 from ..configs.datasets.montgomery.default import dataset 

64 

65 train_samples_weights = get_samples_weights(dataset['__train__']).numpy() 

66 

67 unique, counts = np.unique(train_samples_weights, return_counts=True) 

68 

69 np.testing.assert_equal(counts, np.array([51, 37])) 

70 np.testing.assert_equal(unique, np.array(1 / counts, dtype=np.float32)) 

71 

72@pytest.mark.skip_if_rc_var_not_set('bob.med.tb.nih_cxr14_re.datadir') 

73def test_get_samples_weights_multi(): 

74 

75 from ..configs.datasets.nih_cxr14_re.default import dataset 

76 

77 train_samples_weights = get_samples_weights(dataset['__train__']).numpy() 

78 

79 np.testing.assert_equal( 

80 train_samples_weights, 

81 np.ones(len(dataset['__train__'])) 

82 ) 

83 

84def test_get_samples_weights_concat(): 

85 

86 # Temporarily modify Montgomery datadir 

87 new_value = {"bob.med.tb.montgomery.datadir": montgomery_datadir} 

88 with rc_context(**new_value): 

89 

90 from ..configs.datasets.montgomery.default import dataset 

91 

92 train_dataset = ConcatDataset((dataset['__train__'], dataset['__train__'])) 

93 

94 train_samples_weights = get_samples_weights(train_dataset).numpy() 

95 

96 unique, counts = np.unique(train_samples_weights, return_counts=True) 

97 

98 np.testing.assert_equal(counts, np.array([102, 74])) 

99 np.testing.assert_equal(unique, np.array(2 / counts, dtype=np.float32)) 

100 

101@pytest.mark.skip_if_rc_var_not_set('bob.med.tb.nih_cxr14_re.datadir') 

102def test_get_samples_weights_multi_concat(): 

103 

104 from ..configs.datasets.nih_cxr14_re.default import dataset 

105 

106 train_dataset = ConcatDataset((dataset['__train__'], dataset['__train__'])) 

107 

108 train_samples_weights = get_samples_weights(train_dataset).numpy() 

109 

110 ref_samples_weights = np.concatenate(( 

111 torch.full((len(dataset['__train__']),), 1. / len(dataset['__train__'])), 

112 torch.full((len(dataset['__train__']),), 1. / len(dataset['__train__'])), 

113 )) 

114 

115 np.testing.assert_equal(train_samples_weights, ref_samples_weights) 

116 

117def test_get_positive_weights(): 

118 

119 # Temporarily modify Montgomery datadir 

120 new_value = {"bob.med.tb.montgomery.datadir": montgomery_datadir} 

121 with rc_context(**new_value): 

122 

123 from ..configs.datasets.montgomery.default import dataset 

124 

125 train_positive_weights = get_positive_weights(dataset['__train__']).numpy() 

126 

127 np.testing.assert_equal( 

128 train_positive_weights, 

129 np.array([51.0/37.0], 

130 dtype=np.float32) 

131 ) 

132 

133@pytest.mark.skip_if_rc_var_not_set('bob.med.tb.nih_cxr14_re.datadir') 

134def test_get_positive_weights_multi(): 

135 

136 from ..configs.datasets.nih_cxr14_re.default import dataset 

137 

138 train_positive_weights = get_positive_weights(dataset['__train__']).numpy() 

139 valid_positive_weights = get_positive_weights(dataset['__valid__']).numpy() 

140 

141 assert torch.all( 

142 torch.eq( 

143 torch.FloatTensor(np.around(train_positive_weights, 4)), 

144 torch.FloatTensor( 

145 np.around( 

146 [ 

147 0.9195434, 

148 0.9462068, 

149 0.8070095, 

150 0.94879204, 

151 0.767055, 

152 0.8944615, 

153 0.88212335, 

154 0.8227136, 

155 0.8943905, 

156 0.8864118, 

157 0.90026057, 

158 0.8888551, 

159 0.884739, 

160 0.84540284, 

161 ], 

162 4, 

163 ) 

164 ), 

165 ) 

166 ) 

167 

168 assert torch.all( 

169 torch.eq( 

170 torch.FloatTensor(np.around(valid_positive_weights, 4)), 

171 torch.FloatTensor( 

172 np.around( 

173 [ 

174 0.9366929, 

175 0.9535433, 

176 0.79543304, 

177 0.9530709, 

178 0.74834645, 

179 0.88708663, 

180 0.86661416, 

181 0.81496066, 

182 0.89480317, 

183 0.8888189, 

184 0.8933858, 

185 0.89795274, 

186 0.87181103, 

187 0.8266142, 

188 ], 

189 4, 

190 ) 

191 ), 

192 ) 

193 ) 

194 

195def test_get_positive_weights_concat(): 

196 

197 # Temporarily modify Montgomery datadir 

198 new_value = {"bob.med.tb.montgomery.datadir": montgomery_datadir} 

199 with rc_context(**new_value): 

200 

201 from ..configs.datasets.montgomery.default import dataset 

202 

203 train_dataset = ConcatDataset((dataset['__train__'], dataset['__train__'])) 

204 

205 train_positive_weights = get_positive_weights(train_dataset).numpy() 

206 

207 np.testing.assert_equal( 

208 train_positive_weights, 

209 np.array([51.0/37.0], 

210 dtype=np.float32) 

211 ) 

212 

213@pytest.mark.skip_if_rc_var_not_set('bob.med.tb.nih_cxr14_re.datadir') 

214def test_get_positive_weights_multi_concat(): 

215 

216 from ..configs.datasets.nih_cxr14_re.default import dataset 

217 

218 train_dataset = ConcatDataset((dataset['__train__'], dataset['__train__'])) 

219 valid_dataset = ConcatDataset((dataset['__valid__'], dataset['__valid__'])) 

220 

221 train_positive_weights = get_positive_weights(train_dataset).numpy() 

222 valid_positive_weights = get_positive_weights(valid_dataset).numpy() 

223 

224 assert torch.all( 

225 torch.eq( 

226 torch.FloatTensor(np.around(train_positive_weights, 4)), 

227 torch.FloatTensor( 

228 np.around( 

229 [ 

230 0.9195434, 

231 0.9462068, 

232 0.8070095, 

233 0.94879204, 

234 0.767055, 

235 0.8944615, 

236 0.88212335, 

237 0.8227136, 

238 0.8943905, 

239 0.8864118, 

240 0.90026057, 

241 0.8888551, 

242 0.884739, 

243 0.84540284, 

244 ], 

245 4, 

246 ) 

247 ), 

248 ) 

249 ) 

250 

251 assert torch.all( 

252 torch.eq( 

253 torch.FloatTensor(np.around(valid_positive_weights, 4)), 

254 torch.FloatTensor( 

255 np.around( 

256 [ 

257 0.9366929, 

258 0.9535433, 

259 0.79543304, 

260 0.9530709, 

261 0.74834645, 

262 0.88708663, 

263 0.86661416, 

264 0.81496066, 

265 0.89480317, 

266 0.8888189, 

267 0.8933858, 

268 0.89795274, 

269 0.87181103, 

270 0.8266142, 

271 ], 

272 4, 

273 ) 

274 ), 

275 ) 

276 )