File size: 7,789 Bytes
4b7c478
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
import math
import torch
import torch.utils.data
from pathlib import Path
from torchvision import datasets, transforms
import multiprocessing

from .helpers import compute_mean_and_std, get_data_location
import matplotlib.pyplot as plt


def get_data_loaders(
    batch_size: int = 32, valid_size: float = 0.2, num_workers: int = -1, limit: int = -1
):
    """
    Create and returns the train_one_epoch, validation and test data loaders.

    :param batch_size: size of the mini-batches
    :param valid_size: fraction of the dataset to use for validation. For example 0.2
                       means that 20% of the dataset will be used for validation
    :param num_workers: number of workers to use in the data loaders. Use -1 to mean
                        "use all my cores"
    :param limit: maximum number of data points to consider
    :return a dictionary with 3 keys: 'train_one_epoch', 'valid' and 'test' containing respectively the
            train_one_epoch, validation and test data loaders
    """

    if num_workers == -1:
        # Use all cores
        num_workers = multiprocessing.cpu_count()

    # We will fill this up later
    data_loaders = {"train": None, "valid": None, "test": None}

    base_path = Path(get_data_location())

    # Compute mean and std of the dataset
    mean, std = compute_mean_and_std()
    print(f"Dataset mean: {mean}, std: {std}")

    # MY CODE HERE:
    # create 3 sets of data transforms: one for the training dataset,
    # containing data augmentation, one for the validation dataset
    # (without data augmentation) and one for the test set (again
    # without augmentation)
    # HINT: resize the image to 256 first, then crop them to 224, then add the
    # appropriate transforms for that step
    data_transforms = {
        "train": transforms.Compose(
            [transforms.Resize(256),
            transforms.RandomResizedCrop(224, scale=(0.8,1.0)),
            transforms.RandAugment(2),
            transforms.RandomHorizontalFlip(),
            transforms.ToTensor(),
            transforms.Normalize(mean.tolist(),std.tolist())]
        ),
        "valid": transforms.Compose(
            [transforms.Resize((224,224)),
            transforms.ToTensor(),
            transforms.Normalize(mean.tolist(),std.tolist())]
        ),
        "test": transforms.Compose(
            [transforms.Resize((224,224)),
            transforms.ToTensor(),
            transforms.Normalize(mean.tolist(),std.tolist())]
        ),
    }

    # Create train and validation datasets
    train_data = datasets.ImageFolder(
        base_path / "train",
        # MY CODE HERE: add the appropriate transform that you defined in   
        # the data_transforms dictionary
        data_transforms["train"]
    )
    # The validation dataset is a split from the train_one_epoch dataset, so we read
    # from the same folder, but we apply the transforms for validation
    valid_data = datasets.ImageFolder(
        base_path / "train",
        # MY CODE HERE: add the appropriate transform that you defined in
        # the data_transforms dictionary
        data_transforms["valid"]
    )

    # obtain training indices that will be used for validation
    n_tot = len(train_data)
    indices = torch.randperm(n_tot)

    # If requested, limit the number of data points to consider
    if limit > 0:
        indices = indices[:limit]
        n_tot = limit

    split = int(math.ceil(valid_size * n_tot))
    train_idx, valid_idx = indices[split:], indices[:split]

    # define samplers for obtaining training and validation batches
    train_sampler = torch.utils.data.SubsetRandomSampler(train_idx)
    valid_sampler  = torch.utils.data.SubsetRandomSampler(valid_idx)

    # prepare data loaders
    data_loaders["train"] = torch.utils.data.DataLoader(
        train_data,
        batch_size=batch_size,
        sampler=train_sampler,
        num_workers=num_workers,
    )
    data_loaders["valid"] = torch.utils.data.DataLoader(
        # MY CODE HERE
        valid_data,
        batch_size=batch_size,
        sampler=valid_sampler,
        num_workers=num_workers,
    )

    # Now create the test data loader
    test_data = datasets.ImageFolder(
        base_path / "test",
        # MY CODE HERE (add the test transform)
        data_transforms["test"]
    )

    if limit > 0:
        indices = torch.arange(limit)
        test_sampler = torch.utils.data.SubsetRandomSampler(indices)
    else:
        test_sampler = None

    data_loaders["test"] = torch.utils.data.DataLoader(
        # MY CODE HERE (remember to add shuffle=False as well)
        test_data,
        batch_size=batch_size,
        sampler=test_sampler,
        num_workers=num_workers,
        shuffle=False,
    )

    return data_loaders


def visualize_one_batch(data_loaders, max_n: int = 5):
    """
    Visualize one batch of data.

    :param data_loaders: dictionary containing data loaders
    :param max_n: maximum number of images to show
    :return: None
    """

    # MY CODE HERE:
    # obtain one batch of training images
    # First obtain an iterator from the train dataloader
    dataiter  = iter(data_loaders["train"])
    # Then call the .next() method on the iterator you just
    # obtained (iter.next() is deprecated, using next(iter)).
    images, labels  = next(dataiter)

    # Undo the normalization (for visualization purposes)
    mean, std = compute_mean_and_std()
    invTrans = transforms.Compose(
        [
            transforms.Normalize(mean=[0.0, 0.0, 0.0], std=1 / std),
            transforms.Normalize(mean=-mean, std=[1.0, 1.0, 1.0]),
        ]
    )

    images = invTrans(images)

    # MY CODE HERE:
    # Get class names from the train data loader
    class_names  = data_loaders["train"].dataset.classes

    # Convert from BGR (the format used by pytorch) to
    # RGB (the format expected by matplotlib)
    images = torch.permute(images, (0, 2, 3, 1)).clip(0, 1)

    # plot the images in the batch, along with the corresponding labels
    fig = plt.figure(figsize=(25, 4))
    for idx in range(max_n):
        ax = fig.add_subplot(1, max_n, idx + 1, xticks=[], yticks=[])
        ax.imshow(images[idx])
        # print out the correct label for each image
        # .item() gets the value contained in a Tensor
        ax.set_title(class_names[labels[idx].item()])


######################################################################################
#                                     TESTS
######################################################################################
import pytest


@pytest.fixture(scope="session")
def data_loaders():
    return get_data_loaders(batch_size=2, num_workers=0)


def test_data_loaders_keys(data_loaders):

    assert set(data_loaders.keys()) == {"train", "valid", "test"}, "The keys of the data_loaders dictionary should be train, valid and test"


def test_data_loaders_output_type(data_loaders):
    # Test the data loaders
    dataiter = iter(data_loaders["train"])
    images, labels = next(dataiter)

    assert isinstance(images, torch.Tensor), "images should be a Tensor"
    assert isinstance(labels, torch.Tensor), "labels should be a Tensor"
    assert images[0].shape[-1] == 224, "The tensors returned by your dataloaders should be 224x224. Did you " \
                                       "forget to resize and/or crop?"


def test_data_loaders_output_shape(data_loaders):
    dataiter = iter(data_loaders["train"])
    images, labels = next(dataiter)

    assert len(images) == 2, f"Expected a batch of size 2, got size {len(images)}"
    assert (
        len(labels) == 2
    ), f"Expected a labels tensor of size 2, got size {len(labels)}"


def test_visualize_one_batch(data_loaders):

    visualize_one_batch(data_loaders, max_n=2)