Downsample Binary Pytorch Dataset Down To Size Of Smallest Class

downsample.py

import torch
from torch.utils.data import ConcatDataset, DataLoader, Dataset, Subset
import torchvision
import numpy as np
import matplotlib.pyplot as plt
import time
import os
from PIL import Image
from tempfile import TemporaryDirectory
import random

def downsample_balance_binary_dataset(dataset: Dataset) -> Dataset:
    class_0_count = len([d for d in dataset if d[1] == 0])
    class_1_count = len([d for d in dataset if d[1] == 1])
    if class_0_count > class_1_count:
        class_0_indeces = [i for i, val in enumerate(dataset) if val[1] == 0]
        class_1_indeces = [i for i, val in enumerate(dataset) if val[1] == 1]
        downsampled_class_0_indeces = class_0_indeces[:class_1_count]
        if len(class_1_indeces) != len(downsampled_class_0_indeces):
            raise ValueError("Error during downsampling, class_1_indices was not the same as downsampled_class_0_indeces")
        all_indices = downsampled_class_0_indeces + class_1_indeces
        print(f"Sampled dataset down to {len(all_indices)} samples")
        return Subset(dataset, all_indices)
    elif class_1_count > class_0_count:
        class_0_indeces = [i for i, val in enumerate(dataset) if val[1] == 0]
        class_1_indeces = [i for i, val in enumerate(dataset) if val[1] == 1]
        downsampled_class_1_indeces = class_1_indeces[:class_0_count]
        all_indices = downsampled_class_1_indeces + class_0_indeces
        if len(class_0_indeces) != len(downsampled_class_1_indeces):
            raise ValueError("Error during downsampling, class_0_indices was not the same as downsampled_class_0_indeces")
        print(f"Sampled dataset down to {len(all_indices)} samples")
        return Subset(dataset, all_indices)
    else:
        return dataset