cnn0 / DATA.md
shyo2's picture
Devin
Add results writeup, data provenance, and README
ee3b201
|
Raw History Blame Contribute Delete
1.14 kB
# 데이터 좜처 및 λ‹€μš΄λ‘œλ“œ 방법
## MNIST
- **λ‚΄μš©**: 28Γ—28 흑백 손글씨 숫자 이미지, 10클래슀 (0~9)
- **원본 좜처**: http://yann.lecun.com/exdb/mnist/
- `train-images-idx3-ubyte.gz` / `train-labels-idx1-ubyte.gz` (60,000 train)
- `t10k-images-idx3-ubyte.gz` / `t10k-labels-idx1-ubyte.gz` (10,000 test)
- **λΌμ΄μ„ μŠ€**: CC BY-SA 3.0 (원본 μ‚¬μ΄νŠΈ κΈ°μ€€)
### λ‹€μš΄λ‘œλ“œ 방법
μ½”λ“œμ—μ„œ `torchvision.datasets.MNIST`λ₯Ό μ‚¬μš©ν•˜λ©°, 졜초 μ‹€ν–‰ μ‹œ 원본 μ‚¬μ΄νŠΈμ—μ„œ
μžλ™μœΌλ‘œ λ‹€μš΄λ‘œλ“œλœλ‹€. 별도 μˆ˜λ™ λ‹€μš΄λ‘œλ“œ λΆˆν•„μš”.
```python
# src/train.py λ‚΄ get_loaders()
train_set = datasets.MNIST(data_dir, train=True, download=True, transform=transform)
val_set = datasets.MNIST(data_dir, train=False, download=True, transform=transform)
```
- μ €μž₯ μœ„μΉ˜: `data/MNIST/raw/` (`.gitignore`둜 μ œμ™Έ, 리포에 ν¬ν•¨ν•˜μ§€ μ•ŠμŒ)
- μ „μ²˜λ¦¬: `ToTensor()` β†’ `Normalize((0.1307,), (0.3081,))` (MNIST 전체 λ°μ΄ν„°μ…‹μ˜ mean/std)
## μž¬ν˜„ λͺ…λ Ή
```bash
./.venv/Scripts/python src/train.py --depth 20 --epochs 15 # 졜초 μ‹€ν–‰ μ‹œ data/ μžλ™ 생성
```