105. PyTorch: Dataset & DataLoader

Implement a custom Dataset and use DataLoader to produce batches. Return the size of each batch to verify the batching behavior.

Signature: def create_batches(data, labels, batch_size, shuffle=False)

• data: list of feature vectors (list of lists)
• labels: list of ints
• batch_size: int
• shuffle: bool (use False for deterministic tests)
• Returns: list of ints — the number of samples in each batch

Implement SimpleDataset(Dataset) with:

• __init__(self, data, labels): store as float32 / long tensors
• __len__(self): return dataset size
• __getitem__(self, idx): return (data[idx], labels[idx])

Then build DataLoader(dataset, batch_size=batch_size, shuffle=shuffle) and return [len(bx) for bx, _ in loader].

Why? DataLoader handles batching, shuffling, and worker processes — core infrastructure for any training loop.