Hyperdimensional-Computing/utils.py at master · Cornell-RelaxML/Hyperdimensional-Computing · GitHub

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
#!/usr/bin/env python
# coding: utf-8
import torch
import torchvision
import numpy as np
import time
from torchvision import transforms
import matplotlib.pyplot as plt
from encoder import LinearEncoder, RandomFourierEncoder


def quantize(data, precision=8):
    # assume min and max of the data is -1 and 1
    scaling_factor = 2 ** (precision - 1) - 1
    data = np.round(data * scaling_factor)
    # shift the quantized data to positive and rescale to [0, 1.0]
    return (data + scaling_factor) / 255.0


def encode_and_save(args):
    ### load data using torch with pixel values in [0,1]
    transform = transforms.Compose([
        transforms.ToTensor(),
    ])
    if args.dataset == 'mnist':
        trainset = torchvision.datasets.MNIST(root=args.raw_data_dir, train=True, download=True, transform=transform)
        testset = torchvision.datasets.MNIST(root=args.raw_data_dir, train=False, download=True, transform=transform)
    elif args.dataset == 'fmnist':
        trainset = torchvision.datasets.FashionMNIST(root=args.raw_data_dir, train=True, download=True,
                                                     transform=transform)
        testset = torchvision.datasets.FashionMNIST(root=args.raw_data_dir, train=False, download=True,
                                                    transform=transform)
    elif args.dataset == 'cifar':
        trainset = torchvision.datasets.CIFAR10(root=args.raw_data_dir, train=True, download=True, transform=transform)
        testset = torchvision.datasets.CIFAR10(root=args.raw_data_dir, train=False, download=True, transform=transform)
    elif args.dataset == "isolet":
        import pickle
        def dataset(source):
            with open(source, 'rb') as f:
                isolet = pickle.load(f)
            trainData, trainLabels, testData, testLabels = isolet
            return np.array(trainData), np.array(trainLabels), np.array(testData), np.array(testLabels)

        x_train, y_train, x_test, y_test = dataset(source=f'./{args.raw_data_dir}/isolet/isolet.pkl')
        x_train, y_train = torch.tensor(quantize(x_train, precision=8)).unsqueeze(1), torch.tensor(y_train).long()
        x_test, y_test = torch.tensor(quantize(x_test, precision=8)).unsqueeze(1), torch.tensor(y_test).long()
        trainset = HDDataset(x_train, y_train)
        testset = HDDataset(x_test, y_test)
    elif args.dataset == "ucihar":
        x_train_path = f'./{args.raw_data_dir}/ucihar/train/x_train.txt'
        y_train_path = f'./{args.raw_data_dir}/ucihar/train/y_train.txt'
        x_test_path = f'./{args.raw_data_dir}/ucihar/test/x_test.txt'
        y_test_path = f'./{args.raw_data_dir}/ucihar/test/y_test.txt'

        def load_data(feature_file_path, label_file_path):
            # load training features from txt
            x_train = open(feature_file_path, 'r')
            x_train = x_train.readlines()
            for idx in range(len(x_train)):
                x_train[idx] = x_train[idx].split()
            x_train = np.array(x_train, dtype=np.float32)
            # load test features from txt
            y_train = open(label_file_path, 'r')
            y_train = y_train.readlines()
            y_train = np.array(y_train, dtype=np.int32) - 1
            return x_train, y_train

        x_train, y_train = load_data(x_train_path, y_train_path)
        x_test, y_test = load_data(x_test_path, y_test_path)
        x_train, y_train = torch.tensor(quantize(x_train, precision=8)).unsqueeze(1), torch.tensor(y_train).long()
        x_test, y_test = torch.tensor(quantize(x_test, precision=8)).unsqueeze(1), torch.tensor(y_test).long()
        trainset = HDDataset(x_train, y_train)
        testset = HDDataset(x_test, y_test)
    else:
        raise ValueError("Dataset is not supported.")
    assert len(trainset[0][0].size()) > 1
    channels = trainset[0][0].size(0)
    print('# of channels of data', channels)
    input_dim = torch.prod(torch.tensor(list(trainset[0][0].size())))
    print('# of training samples and test samples', len(trainset), len(testset))

    if args.model == 'linear-hdc':
        print("Encoding to binary HDC with linear hamming distance.")
        encoder = LinearEncoder(dim=args.dim)
    elif 'rff' in args.model:
        print("Encoding with random fourier features encoder.")
        encoder = RandomFourierEncoder(
            input_dim=input_dim, gamma=args.gamma, gorder=args.gorder, output_dim=args.dim)
    else:
        raise ValueError("No such feature type is supported.")

    mem = encoder.build_item_mem()
    print("Encoded pixels to hypervectors with size: ", mem.size())
    torch.save(mem, f'{args.data_dir}/item_mem.pt')

    print("Encoding training data...")
    train_hd, y_train = encoder.encode_data_extract_labels(trainset)
    torch.save(train_hd, f'{args.data_dir}/train_hd.pt')
    torch.save(y_train, f'{args.data_dir}/y_train.pt')
    del train_hd, y_train
    torch.cuda.empty_cache()  # in case of CUDA OOM

    print("Encoding test data...")
    test_hd, y_test = encoder.encode_data_extract_labels(testset)
    torch.save(test_hd, f'{args.data_dir}/test_hd.pt')
    torch.save(y_test, f'{args.data_dir}/y_test.pt')
    del test_hd, y_test
    torch.cuda.empty_cache()


def load(args):
    #     mem = torch.load(f'{args.data_dir}/item_mem.pt')
    #     print("Loaded pixel hypervectors with size: ", mem.size())

    print("Loading encoded training data...")
    train_hd = torch.load(f'{args.data_dir}/train_hd.pt')
    y_train = torch.load(f'{args.data_dir}/y_train.pt')

    print("Loading encoded test data...")
    test_hd = torch.load(f'{args.data_dir}/test_hd.pt')
    y_test = torch.load(f'{args.data_dir}/y_test.pt')

    print(f"Size of encoded training data {train_hd.size()} and test data {test_hd.size()}")
    return train_hd, y_train, test_hd, y_test


class HDDataset(torch.utils.data.Dataset):
    def __init__(self, data, labels):
        self.data = data
        self.labels = labels

    def __len__(self):
        return self.data.size(0)

    def __getitem__(self, idx):
        return self.data[idx], self.labels[idx]


def prepare_data(args):
    train_hd, y_train, test_hd, y_test = load(args)
    train_dataset = HDDataset(train_hd, y_train)
    test_dataset = HDDataset(test_hd, y_test)
    trainloader = torch.utils.data.DataLoader(train_dataset,
                                              batch_size=16,
                                              shuffle=True,
                                              pin_memory=True,
                                              num_workers=1)
    testloader = torch.utils.data.DataLoader(test_dataset,
                                             batch_size=16,
                                             shuffle=False,
                                             pin_memory=True,
                                             num_workers=1)
    return trainloader, testloader