dl-env

[project]
name = "dl-env"
version = "0.1.0"
description = "Deep learning baseline environment"
requires-python = "==3.12.*"
dependencies = [
  "torch==2.10.0",
  "torchvision==0.25.0",
  "torchaudio==2.10.0",
  "scikit-learn==1.8.0",
  "xgboost==3.2.0",
  "lightgbm==4.6.0",
  "statsmodels==0.14.6",
  "shap==0.50.0",
  "mlflow==3.10.0",
  "gradio==6.6.0",
  "jupyterlab==4.5.4",
  "ipykernel==7.2.0",
  "pytest==9.0.2",
  "ruff==0.15.2",
]

[tool.uv]
package = false

uv run python -c "import torch, sklearn, xgboost, lightgbm, shap; print('torch', torch.__version__); print('cuda', torch.cuda.is_available()); print('sklearn', sklearn.__version__)"

import torch
import torch.nn as nn
from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
import shap
import mlflow

# 1) 데이터 준비
X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
scaler = StandardScaler().fit(X_train)
X_train, X_test = scaler.transform(X_train), scaler.transform(X_test)

# 2) 모델 정의 및 학습
model = nn.Sequential(nn.Linear(4, 32), nn.ReLU(), nn.Linear(32, 3))
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
criterion = nn.CrossEntropyLoss()

for epoch in range(50):
    out = model(torch.tensor(X_train, dtype=torch.float32))
    loss = criterion(out, torch.tensor(y_train))
    optimizer.zero_grad(); loss.backward(); optimizer.step()

# 3) 평가
with torch.no_grad():
    preds = model(torch.tensor(X_test, dtype=torch.float32)).argmax(dim=1)
    acc = (preds.numpy() == y_test).mean()

# 4) SHAP 해석
explainer = shap.Explainer(lambda x: model(torch.tensor(x, dtype=torch.float32)).detach().numpy(), X_train)
shap_values = explainer(X_test[:10])

# 5) MLflow 기록
with mlflow.start_run():
    mlflow.log_param("epochs", 50)
    mlflow.log_metric("accuracy", acc)
    mlflow.pytorch.log_model(model, "model")