PFLlib/dataset/generate_Digit5.py

# PFLlib: Personalized Federated Learning Algorithm Library
# Copyright (C) 2021  Jianqing Zhang

# This program is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 2 of the License, or
# (at your option) any later version.

# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.

# You should have received a copy of the GNU General Public License along
# with this program; if not, write to the Free Software Foundation, Inc.,
# 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.

import numpy as np
import os
import random
import torchvision.transforms as transforms
import torch.utils.data as data
from utils.dataset_utils import split_data, save_file
from os import path
from scipy.io import loadmat
from PIL import Image
from torch.utils.data import DataLoader


# https://github.com/FengHZ/KD3A/blob/master/datasets/DigitFive.py
def load_mnist(base_path):
    print("load mnist")
    mnist_data = loadmat(path.join(base_path, "mnist_data.mat"))
    mnist_train = np.reshape(mnist_data['train_32'], (55000, 32, 32, 1))
    mnist_test = np.reshape(mnist_data['test_32'], (10000, 32, 32, 1))
    # turn to the 3 channel image with C*H*W
    mnist_train = np.concatenate([mnist_train, mnist_train, mnist_train], 3)
    mnist_test = np.concatenate([mnist_test, mnist_test, mnist_test], 3)
    mnist_train = mnist_train.transpose(0, 3, 1, 2).astype(np.float32)
    mnist_test = mnist_test.transpose(0, 3, 1, 2).astype(np.float32)
    # get labels
    mnist_labels_train = mnist_data['label_train']
    mnist_labels_test = mnist_data['label_test']
    # random sample 25000 from train dataset and random sample 9000 from test dataset
    train_label = np.argmax(mnist_labels_train, axis=1)
    inds = np.random.permutation(mnist_train.shape[0])
    mnist_train = mnist_train[inds]
    train_label = train_label[inds]
    test_label = np.argmax(mnist_labels_test, axis=1)

    mnist_train = mnist_train[:25000]
    train_label = train_label[:25000]
    mnist_test = mnist_test[:9000]
    test_label = test_label[:9000]
    return mnist_train, train_label, mnist_test, test_label


def load_mnist_m(base_path):
    print("load mnist_m")
    mnistm_data = loadmat(path.join(base_path, "mnistm_with_label.mat"))
    mnistm_train = mnistm_data['train']
    mnistm_test = mnistm_data['test']
    mnistm_train = mnistm_train.transpose(0, 3, 1, 2).astype(np.float32)
    mnistm_test = mnistm_test.transpose(0, 3, 1, 2).astype(np.float32)
    # get labels
    mnistm_labels_train = mnistm_data['label_train']
    mnistm_labels_test = mnistm_data['label_test']
    # random sample 25000 from train dataset and random sample 9000 from test dataset
    train_label = np.argmax(mnistm_labels_train, axis=1)
    inds = np.random.permutation(mnistm_train.shape[0])
    mnistm_train = mnistm_train[inds]
    train_label = train_label[inds]
    test_label = np.argmax(mnistm_labels_test, axis=1)
    mnistm_train = mnistm_train[:25000]
    train_label = train_label[:25000]
    mnistm_test = mnistm_test[:9000]
    test_label = test_label[:9000]
    return mnistm_train, train_label, mnistm_test, test_label


def load_svhn(base_path):
    print("load svhn")
    svhn_train_data = loadmat(path.join(base_path, "svhn_train_32x32.mat"))
    svhn_test_data = loadmat(path.join(base_path, "svhn_test_32x32.mat"))
    svhn_train = svhn_train_data['X']
    svhn_train = svhn_train.transpose(3, 2, 0, 1).astype(np.float32)
    svhn_test = svhn_test_data['X']
    svhn_test = svhn_test.transpose(3, 2, 0, 1).astype(np.float32)
    train_label = svhn_train_data["y"].reshape(-1)
    test_label = svhn_test_data["y"].reshape(-1)
    inds = np.random.permutation(svhn_train.shape[0])
    svhn_train = svhn_train[inds]
    train_label = train_label[inds]
    svhn_train = svhn_train[:25000]
    train_label = train_label[:25000]
    svhn_test = svhn_test[:9000]
    test_label = test_label[:9000]
    train_label[train_label == 10] = 0
    test_label[test_label == 10] = 0
    return svhn_train, train_label, svhn_test, test_label


def load_syn(base_path):
    print("load syn")
    syn_train_data = loadmat(path.join(base_path, "synth_train_32x32.mat"))
    syn_test_data = loadmat(path.join(base_path, "synth_test_32x32.mat"))
    syn_train = syn_train_data["X"]
    syn_test = syn_test_data["X"]
    syn_train = syn_train.transpose(3, 2, 0, 1).astype(np.float32)
    syn_test = syn_test.transpose(3, 2, 0, 1).astype(np.float32)
    train_label = syn_train_data["y"].reshape(-1)
    test_label = syn_test_data["y"].reshape(-1)
    syn_train = syn_train[:25000]
    syn_test = syn_test[:9000]
    train_label = train_label[:25000]
    test_label = test_label[:9000]
    train_label[train_label == 10] = 0
    test_label[test_label == 10] = 0
    return syn_train, train_label, syn_test, test_label


def load_usps(base_path):
    print("load usps")
    usps_dataset = loadmat(path.join(base_path, "usps_28x28.mat"))
    usps_dataset = usps_dataset["dataset"]
    usps_train = usps_dataset[0][0]
    train_label = usps_dataset[0][1]
    train_label = train_label.reshape(-1)
    train_label[train_label == 10] = 0
    usps_test = usps_dataset[1][0]
    test_label = usps_dataset[1][1]
    test_label = test_label.reshape(-1)
    test_label[test_label == 10] = 0
    usps_train = usps_train * 255
    usps_test = usps_test * 255
    usps_train = np.concatenate([usps_train, usps_train, usps_train], 1)
    usps_train = np.tile(usps_train, (4, 1, 1, 1))
    train_label = np.tile(train_label,4)
    usps_train = usps_train[:25000]
    train_label = train_label[:25000]
    usps_test = np.concatenate([usps_test, usps_test, usps_test], 1)
    return usps_train, train_label, usps_test, test_label

class Digit5Dataset(data.Dataset):
    def __init__(self, data, labels, transform=None, target_transform=None):
        super(Digit5Dataset, self).__init__()
        self.data = data
        self.labels = labels
        self.transform = transform
        self.target_transform = target_transform

    def __getitem__(self, index):
        img, label = self.data[index], self.labels[index]
        if img.shape[0] != 1:
            # transpose to Image type,so that the transform function can be used
            img = Image.fromarray(np.uint8(np.asarray(img.transpose((1, 2, 0)))))

        elif img.shape[0] == 1:
            im = np.uint8(np.asarray(img))
            # turn the raw image into 3 channels
            im = np.vstack([im, im, im]).transpose((1, 2, 0))
            img = Image.fromarray(im)

        # do transform with PIL
        if self.transform is not None:
            img = self.transform(img)
        if self.target_transform is not None:
            label = self.target_transform(label)
        return img, label

    def __len__(self):
        return self.data.shape[0]

def digit5_dataset_read(base_path, domain):
    if domain == "mnist":
        train_image, train_label, test_image, test_label = load_mnist(base_path)
    elif domain == "mnistm":
        train_image, train_label, test_image, test_label = load_mnist_m(base_path)
    elif domain == "svhn":
        train_image, train_label, test_image, test_label = load_svhn(base_path)
    elif domain == "syn":
        train_image, train_label, test_image, test_label = load_syn(base_path)
    elif domain == "usps":
        train_image, train_label, test_image, test_label = load_usps(base_path)
    else:
        raise NotImplementedError("Domain {} Not Implemented".format(domain))
    # define the transform function
    transform = transforms.Compose([
        transforms.Resize(32),
        transforms.ToTensor(),
        transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
    ])
    # raise train and test data loader
    train_dataset = Digit5Dataset(data=train_image, labels=train_label, transform=transform)
    train_loader = DataLoader(dataset=train_dataset, batch_size=len(train_dataset), shuffle=False)
    test_dataset = Digit5Dataset(data=test_image, labels=test_label, transform=transform)
    test_loader = DataLoader(dataset=test_dataset, batch_size=len(test_dataset), shuffle=False)
    return train_loader, test_loader


random.seed(1)
np.random.seed(1)
data_path = "Digit5/"
dir_path = "Digit5/"

# Allocate data to usersz``
def generate_dataset(dir_path):
    if not os.path.exists(dir_path):
        os.makedirs(dir_path)
        
    # Setup directory for train/test data
    config_path = dir_path + "config.json"
    train_path = dir_path + "train/"
    test_path = dir_path + "test/"

    if not os.path.exists(train_path):
        os.makedirs(train_path)
    if not os.path.exists(test_path):
        os.makedirs(test_path)

    root = data_path+"rawdata"
    
    # Get Digit5 data
    if not os.path.exists(root):
        os.makedirs(root)
        os.system(f'wget https://drive.google.com/u/0/uc?id=1PT6K-_wmsUEUCxoYzDy0mxF-15tvb2Eu&export=download -P {root}')
        os.system(f'unzip {root}/Digit5.zip -d {root}')

    X, y = [], []
    domains = ['mnistm', 'mnist', 'syn', 'usps', 'svhn']
    for d in domains:
        train_loader, test_loader = digit5_dataset_read(root, d)

        for _, tt in enumerate(train_loader):
            train_data, train_label = tt
        for _, tt in enumerate(test_loader):
            test_data, test_label = tt

        dataset_image = []
        dataset_label = []

        dataset_image.extend(train_data.cpu().detach().numpy())
        dataset_image.extend(test_data.cpu().detach().numpy())
        dataset_label.extend(train_label.cpu().detach().numpy())
        dataset_label.extend(test_label.cpu().detach().numpy())

        X.append(np.array(dataset_image))
        y.append(np.array(dataset_label))
    

    labelss = []
    for yy in y:
        labelss.append(len(set(yy)))
    num_clients = len(y)
    print(f'Number of labels: {labelss}')
    print(f'Number of clients: {num_clients}')

    statistic = [[] for _ in range(num_clients)]
    for client in range(num_clients):
        for i in np.unique(y[client]):
            statistic[client].append((int(i), int(sum(y[client]==i))))


    train_data, test_data = split_data(X, y)
    save_file(config_path, train_path, test_path, train_data, test_data, num_clients, max(labelss), 
        statistic, None, None, None)


if __name__ == "__main__":
    generate_dataset(dir_path)