region: ap-northeast-2
env: dev
conf_bucket: gs-retail-awesome-conf-{region}
data_bucket: gs-retail-awesome-data-{region}
output_bucket: gs-retail-awesome-output-{region}
user_id: your_id   # ← 변경
log_level: INFO

model_type: EBM_Regressor
framework: interpretML
hyperparameters:
  max_bins: 256
  interactions: 5
  outer_bags: 8
  random_state: 42
features:
  - MedInc
  - HouseAge
  - AveRooms
  - AveBedrms
  - Population
  - AveOccup
  - Latitude
  - Longitude
target: MedHouseVal
treatment: MedInc  # DoWhy용

import yaml, boto3, s3fs

with open("conf/env.yml")   as f: ENV   = yaml.safe_load(f)
with open("conf/meta.yml")  as f: META  = yaml.safe_load(f)
with open("conf/model.yml") as f: MODEL = yaml.safe_load(f)

USER_ID    = ENV['user_id']
PROJECT    = META['project']
EXPERIMENT = META['experiment']
REGION     = ENV['region']

CONF_PATH   = f"s3://gs-retail-awesome-conf-{REGION}/dev/{USER_ID}/{PROJECT}/{EXPERIMENT}"
DATA_PATH   = f"s3://gs-retail-awesome-data-{REGION}/dev/{USER_ID}/{PROJECT}/v1/data"
OUTPUT_PATH = f"s3://gs-retail-awesome-output-{REGION}/dev/{USER_ID}/{PROJECT}/{EXPERIMENT}"

s3 = boto3.client('s3')
CONF_BUCKET = f"gs-retail-awesome-conf-{REGION}"
for fname in ["env.yml", "meta.yml", "model.yml"]:
    s3.upload_file(f"conf/{fname}", CONF_BUCKET,
        f"dev/{USER_ID}/{PROJECT}/{EXPERIMENT}/{fname}")
    print(f"  ✓ {fname} → conf 버킷")

from sklearn.datasets import fetch_california_housing
from sklearn.model_selection import train_test_split
from interpret.glassbox import ExplainableBoostingRegressor
from sklearn.metrics import mean_squared_error, r2_score
import numpy as np

# ── 데이터 로드
housing = fetch_california_housing()
X = pd.DataFrame(housing.data, columns=housing.feature_names)
y = pd.Series(housing.target, name="MedHouseVal")

# ── 70/15/15 분리
X_train, X_tmp, y_train, y_tmp = train_test_split(X, y, test_size=0.3, random_state=42)
X_val, X_test, y_val, y_test   = train_test_split(X_tmp, y_tmp, test_size=0.5, random_state=42)

# ── EBM 회귀 학습
ebm_reg = ExplainableBoostingRegressor(
    max_bins=256, interactions=5, outer_bags=8, random_state=42)
ebm_reg.fit(X_train, y_train)

# ── 성능 평가
val_pred = ebm_reg.predict(X_val)
val_rmse = np.sqrt(mean_squared_error(y_val, val_pred))
val_r2   = r2_score(y_val, val_pred)
print(f"Val RMSE: {val_rmse:.4f} (~${val_rmse*100:.0f}K)")
print(f"Val R²  : {val_r2:.4f}")

import dowhy
from dowhy import CausalModel

# ── 인과 DAG 정의 (GML 형식)
# MedInc(처치) → MedHouseVal(결과)
# Latitude, Longitude, AveRooms, AveOccup → MedInc AND MedHouseVal (교란)
causal_graph = """
digraph {
    MedInc -> MedHouseVal;
    Latitude -> MedInc; Latitude -> MedHouseVal;
    Longitude -> MedInc; Longitude -> MedHouseVal;
    AveRooms -> MedInc; AveRooms -> MedHouseVal;
    AveOccup -> MedInc; AveOccup -> MedHouseVal;
    HouseAge -> MedHouseVal;
}
"""

model = CausalModel(
    data=pd.concat([X_train, y_train], axis=1),
    treatment="MedInc",
    outcome="MedHouseVal",
    graph=causal_graph
)
print("✓ 인과 모델 생성 완료")

# ── STEP 1: Identify (인과 효과 식별)
identified_estimand = model.identify_effect(proceed_when_unidentifiable=True)
print(identified_estimand)

# ── STEP 2: Estimate (ATE 추정)
estimate = model.estimate_effect(
    identified_estimand,
    method_name="backdoor.linear_regression",
    target_units="ate"
)
print(f"\nATE (MedInc → MedHouseVal): {estimate.value:.4f}")
print(f"해석: MedInc $10K 증가 시 집값 평균 ${estimate.value * 100000:.0f} 변화")

# ── STEP 3: Refute (인과 효과 검증)
refute_random = model.refute_estimate(
    identified_estimand, estimate,
    method_name="random_common_cause"
)
print(refute_random)

refute_placebo = model.refute_estimate(
    identified_estimand, estimate,
    method_name="placebo_treatment_refuter",
    placebo_type="permute"
)
print(refute_placebo)