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""" 16Test Mnist dataset operators 17""" 18import os 19import pytest 20import numpy as np 21import matplotlib.pyplot as plt 22import mindspore.dataset as ds 23import mindspore.dataset.vision.c_transforms as vision 24from mindspore import log as logger 25 26DATA_DIR = "../data/dataset/testMnistData" 27 28 29def load_mnist(path): 30 """ 31 load Mnist data 32 """ 33 labels_path = os.path.join(path, 't10k-labels-idx1-ubyte') 34 images_path = os.path.join(path, 't10k-images-idx3-ubyte') 35 with open(labels_path, 'rb') as lbpath: 36 lbpath.read(8) 37 labels = np.fromfile(lbpath, dtype=np.uint8) 38 with open(images_path, 'rb') as imgpath: 39 imgpath.read(16) 40 images = np.fromfile(imgpath, dtype=np.uint8) 41 images = images.reshape(-1, 28, 28, 1) 42 images[images > 0] = 255 # Perform binarization to maintain consistency with our API 43 return images, labels 44 45 46def visualize_dataset(images, labels): 47 """ 48 Helper function to visualize the dataset samples 49 """ 50 num_samples = len(images) 51 for i in range(num_samples): 52 plt.subplot(1, num_samples, i + 1) 53 plt.imshow(images[i].squeeze(), cmap=plt.cm.gray) 54 plt.title(labels[i]) 55 plt.show() 56 57 58def test_mnist_content_check(): 59 """ 60 Validate MnistDataset image readings 61 """ 62 logger.info("Test MnistDataset Op with content check") 63 data1 = ds.MnistDataset(DATA_DIR, num_samples=100, shuffle=False) 64 images, labels = load_mnist(DATA_DIR) 65 num_iter = 0 66 # in this example, each dictionary has keys "image" and "label" 67 image_list, label_list = [], [] 68 for i, data in enumerate(data1.create_dict_iterator(num_epochs=1, output_numpy=True)): 69 image_list.append(data["image"]) 70 label_list.append("label {}".format(data["label"])) 71 np.testing.assert_array_equal(data["image"], images[i]) 72 np.testing.assert_array_equal(data["label"], labels[i]) 73 num_iter += 1 74 assert num_iter == 100 75 76 77def test_mnist_basic(): 78 """ 79 Validate MnistDataset 80 """ 81 logger.info("Test MnistDataset Op") 82 83 # case 1: test loading whole dataset 84 data1 = ds.MnistDataset(DATA_DIR) 85 num_iter1 = 0 86 for _ in data1.create_dict_iterator(num_epochs=1): 87 num_iter1 += 1 88 assert num_iter1 == 10000 89 90 # case 2: test num_samples 91 data2 = ds.MnistDataset(DATA_DIR, num_samples=500) 92 num_iter2 = 0 93 for _ in data2.create_dict_iterator(num_epochs=1): 94 num_iter2 += 1 95 assert num_iter2 == 500 96 97 # case 3: test repeat 98 data3 = ds.MnistDataset(DATA_DIR, num_samples=200) 99 data3 = data3.repeat(5) 100 num_iter3 = 0 101 for _ in data3.create_dict_iterator(num_epochs=1): 102 num_iter3 += 1 103 assert num_iter3 == 1000 104 105 # case 4: test batch with drop_remainder=False 106 data4 = ds.MnistDataset(DATA_DIR, num_samples=100) 107 assert data4.get_dataset_size() == 100 108 assert data4.get_batch_size() == 1 109 data4 = data4.batch(batch_size=7) # drop_remainder is default to be False 110 assert data4.get_dataset_size() == 15 111 assert data4.get_batch_size() == 7 112 num_iter4 = 0 113 for _ in data4.create_dict_iterator(num_epochs=1): 114 num_iter4 += 1 115 assert num_iter4 == 15 116 117 # case 5: test batch with drop_remainder=True 118 data5 = ds.MnistDataset(DATA_DIR, num_samples=100) 119 assert data5.get_dataset_size() == 100 120 assert data5.get_batch_size() == 1 121 data5 = data5.batch(batch_size=7, drop_remainder=True) # the rest of incomplete batch will be dropped 122 assert data5.get_dataset_size() == 14 123 assert data5.get_batch_size() == 7 124 num_iter5 = 0 125 for _ in data5.create_dict_iterator(num_epochs=1): 126 num_iter5 += 1 127 assert num_iter5 == 14 128 129 130def test_mnist_pk_sampler(): 131 """ 132 Test MnistDataset with PKSampler 133 """ 134 logger.info("Test MnistDataset Op with PKSampler") 135 golden = [0, 0, 0, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 136 5, 5, 5, 6, 6, 6, 7, 7, 7, 8, 8, 8, 9, 9, 9] 137 sampler = ds.PKSampler(3) 138 data = ds.MnistDataset(DATA_DIR, sampler=sampler) 139 num_iter = 0 140 label_list = [] 141 for item in data.create_dict_iterator(num_epochs=1, output_numpy=True): 142 label_list.append(item["label"]) 143 num_iter += 1 144 np.testing.assert_array_equal(golden, label_list) 145 assert num_iter == 30 146 147 148def test_mnist_sequential_sampler(): 149 """ 150 Test MnistDataset with SequentialSampler 151 """ 152 logger.info("Test MnistDataset Op with SequentialSampler") 153 num_samples = 50 154 sampler = ds.SequentialSampler(num_samples=num_samples) 155 data1 = ds.MnistDataset(DATA_DIR, sampler=sampler) 156 data2 = ds.MnistDataset(DATA_DIR, shuffle=False, num_samples=num_samples) 157 label_list1, label_list2 = [], [] 158 num_iter = 0 159 for item1, item2 in zip(data1.create_dict_iterator(num_epochs=1), data2.create_dict_iterator(num_epochs=1)): 160 label_list1.append(item1["label"].asnumpy()) 161 label_list2.append(item2["label"].asnumpy()) 162 num_iter += 1 163 np.testing.assert_array_equal(label_list1, label_list2) 164 assert num_iter == num_samples 165 166 167def test_mnist_exception(): 168 """ 169 Test error cases for MnistDataset 170 """ 171 logger.info("Test error cases for MnistDataset") 172 error_msg_1 = "sampler and shuffle cannot be specified at the same time" 173 with pytest.raises(RuntimeError, match=error_msg_1): 174 ds.MnistDataset(DATA_DIR, shuffle=False, sampler=ds.PKSampler(3)) 175 176 error_msg_2 = "sampler and sharding cannot be specified at the same time" 177 with pytest.raises(RuntimeError, match=error_msg_2): 178 ds.MnistDataset(DATA_DIR, sampler=ds.PKSampler(3), num_shards=2, shard_id=0) 179 180 error_msg_3 = "num_shards is specified and currently requires shard_id as well" 181 with pytest.raises(RuntimeError, match=error_msg_3): 182 ds.MnistDataset(DATA_DIR, num_shards=10) 183 184 error_msg_4 = "shard_id is specified but num_shards is not" 185 with pytest.raises(RuntimeError, match=error_msg_4): 186 ds.MnistDataset(DATA_DIR, shard_id=0) 187 188 error_msg_5 = "Input shard_id is not within the required interval" 189 with pytest.raises(ValueError, match=error_msg_5): 190 ds.MnistDataset(DATA_DIR, num_shards=5, shard_id=-1) 191 with pytest.raises(ValueError, match=error_msg_5): 192 ds.MnistDataset(DATA_DIR, num_shards=5, shard_id=5) 193 with pytest.raises(ValueError, match=error_msg_5): 194 ds.MnistDataset(DATA_DIR, num_shards=2, shard_id=5) 195 196 error_msg_6 = "num_parallel_workers exceeds" 197 with pytest.raises(ValueError, match=error_msg_6): 198 ds.MnistDataset(DATA_DIR, shuffle=False, num_parallel_workers=0) 199 with pytest.raises(ValueError, match=error_msg_6): 200 ds.MnistDataset(DATA_DIR, shuffle=False, num_parallel_workers=256) 201 with pytest.raises(ValueError, match=error_msg_6): 202 ds.MnistDataset(DATA_DIR, shuffle=False, num_parallel_workers=-2) 203 204 error_msg_7 = "Argument shard_id" 205 with pytest.raises(TypeError, match=error_msg_7): 206 ds.MnistDataset(DATA_DIR, num_shards=2, shard_id="0") 207 208 def exception_func(item): 209 raise Exception("Error occur!") 210 211 error_msg_8 = "The corresponding data files" 212 with pytest.raises(RuntimeError, match=error_msg_8): 213 data = ds.MnistDataset(DATA_DIR) 214 data = data.map(operations=exception_func, input_columns=["image"], num_parallel_workers=1) 215 for _ in data.__iter__(): 216 pass 217 with pytest.raises(RuntimeError, match=error_msg_8): 218 data = ds.MnistDataset(DATA_DIR) 219 data = data.map(operations=vision.Decode(), input_columns=["image"], num_parallel_workers=1) 220 data = data.map(operations=exception_func, input_columns=["image"], num_parallel_workers=1) 221 for _ in data.__iter__(): 222 pass 223 with pytest.raises(RuntimeError, match=error_msg_8): 224 data = ds.MnistDataset(DATA_DIR) 225 data = data.map(operations=exception_func, input_columns=["label"], num_parallel_workers=1) 226 for _ in data.__iter__(): 227 pass 228 229 230def test_mnist_visualize(plot=False): 231 """ 232 Visualize MnistDataset results 233 """ 234 logger.info("Test MnistDataset visualization") 235 236 data1 = ds.MnistDataset(DATA_DIR, num_samples=10, shuffle=False) 237 num_iter = 0 238 image_list, label_list = [], [] 239 for item in data1.create_dict_iterator(num_epochs=1, output_numpy=True): 240 image = item["image"] 241 label = item["label"] 242 image_list.append(image) 243 label_list.append("label {}".format(label)) 244 assert isinstance(image, np.ndarray) 245 assert image.shape == (28, 28, 1) 246 assert image.dtype == np.uint8 247 assert label.dtype == np.uint32 248 num_iter += 1 249 assert num_iter == 10 250 if plot: 251 visualize_dataset(image_list, label_list) 252 253 254def test_mnist_usage(): 255 """ 256 Validate MnistDataset image readings 257 """ 258 logger.info("Test MnistDataset usage flag") 259 260 def test_config(usage, mnist_path=None): 261 mnist_path = DATA_DIR if mnist_path is None else mnist_path 262 try: 263 data = ds.MnistDataset(mnist_path, usage=usage, shuffle=False) 264 num_rows = 0 265 for _ in data.create_dict_iterator(num_epochs=1, output_numpy=True): 266 num_rows += 1 267 except (ValueError, TypeError, RuntimeError) as e: 268 return str(e) 269 return num_rows 270 271 assert test_config("test") == 10000 272 assert test_config("all") == 10000 273 assert "MnistDataset API can't read the data file (interface mismatch or no data found)" in test_config("train") 274 assert "usage is not within the valid set of ['train', 'test', 'all']" in test_config("invalid") 275 assert "Argument usage with value ['list'] is not of type [<class 'str'>]" in test_config(["list"]) 276 277 # change this directory to the folder that contains all mnist files 278 all_files_path = None 279 # the following tests on the entire datasets 280 if all_files_path is not None: 281 assert test_config("train", all_files_path) == 60000 282 assert test_config("test", all_files_path) == 10000 283 assert test_config("all", all_files_path) == 70000 284 assert ds.MnistDataset(all_files_path, usage="train").get_dataset_size() == 60000 285 assert ds.MnistDataset(all_files_path, usage="test").get_dataset_size() == 10000 286 assert ds.MnistDataset(all_files_path, usage="all").get_dataset_size() == 70000 287 288 289if __name__ == '__main__': 290 test_mnist_content_check() 291 test_mnist_basic() 292 test_mnist_pk_sampler() 293 test_mnist_sequential_sampler() 294 test_mnist_exception() 295 test_mnist_visualize(plot=True) 296 test_mnist_usage() 297