import sys
sys.path.insert(0, "..")

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

from sklearn.compose import ColumnTransformer
from sklearn.dummy import DummyClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.inspection import permutation_importance
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import (ConfusionMatrixDisplay, RocCurveDisplay, classification_report,
                             f1_score, precision_recall_curve)
from sklearn.model_selection import StratifiedKFold, cross_validate, train_test_split
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OneHotEncoder, StandardScaler
from sklearn.tree import DecisionTreeClassifier

from common.datasets import churn_pelanggan

SEED = 42  # satu seed untuk seluruh proyek → reproducible

df = churn_pelanggan()
TARGET = "churn"
print(df.shape)
df.head()

print("Proporsi churn:", df[TARGET].mean().round(3))
df.describe().T.round(2)

fig, axes = plt.subplots(1, 3, figsize=(15, 3.5))
df.groupby("kontrak")[TARGET].mean().plot.bar(ax=axes[0], title="Churn rate per kontrak", rot=0)
df.boxplot(column="lama_langganan", by=TARGET, ax=axes[1]); axes[1].set_title("Lama langganan vs churn")
df.groupby("jumlah_keluhan")[TARGET].mean().plot(ax=axes[2], marker="o", title="Churn rate vs jumlah keluhan")
plt.suptitle(""); plt.tight_layout(); plt.show()

X, y = df.drop(columns=TARGET), df[TARGET]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.25, stratify=y, random_state=SEED)

num_cols = [c for c in X.columns if pd.api.types.is_numeric_dtype(X[c])]
cat_cols = [c for c in X.columns if c not in num_cols]
print("Numerik    :", num_cols)
print("Kategorikal:", cat_cols)

def bangun_pipeline(model):
    prep = ColumnTransformer([
        ("num", Pipeline([("imputer", SimpleImputer(strategy="median")),
                          ("scaler", StandardScaler())]), num_cols),
        ("cat", Pipeline([("imputer", SimpleImputer(strategy="most_frequent")),
                          ("onehot", OneHotEncoder(handle_unknown="ignore"))]), cat_cols),
    ])
    return Pipeline([("prep", prep), ("model", model)])

bangun_pipeline(LogisticRegression())

models = {
    "baseline (mayoritas)": DummyClassifier(strategy="most_frequent"),
    "decision tree d=5": DecisionTreeClassifier(max_depth=5, random_state=SEED),
    "logreg balanced": LogisticRegression(max_iter=1000, class_weight="balanced"),
    "random forest": RandomForestClassifier(n_estimators=200, class_weight="balanced", random_state=SEED),
}
cv = StratifiedKFold(5, shuffle=True, random_state=SEED)
baris = []
for nama, mdl in models.items():
    r = cross_validate(bangun_pipeline(mdl), X_train, y_train, cv=cv,
                       scoring=["f1", "recall", "precision", "roc_auc"])
    baris.append({"model": nama,
                  **{f"{k[5:]}_mean": v.mean() for k, v in r.items() if k.startswith("test_")},
                  "f1_std": r["test_f1"].std()})
hasil_cv = pd.DataFrame(baris).set_index("model").sort_values("f1_mean", ascending=False)
hasil_cv.round(3)

hasil_cv["f1_mean"].plot.barh(xerr=hasil_cv["f1_std"], figsize=(7, 3), title="F1 CV 5-fold (mean ± std)")
plt.show()

terbaik = hasil_cv.index[0]
pipe = bangun_pipeline(models[terbaik]).fit(X_train, y_train)
pred = pipe.predict(X_test)
print("Model terpilih:", terbaik)
print(classification_report(y_test, pred, digits=3))

fig, axes = plt.subplots(1, 2, figsize=(11, 4))
ConfusionMatrixDisplay.from_predictions(y_test, pred, ax=axes[0], colorbar=False)
RocCurveDisplay.from_estimator(pipe, X_test, y_test, ax=axes[1])
plt.tight_layout(); plt.show()

proba = pipe.predict_proba(X_test)[:, 1]
ambang = np.quantile(proba, 0.80)
pred_kapasitas = (proba >= ambang).astype(int)
print(f"ambang={ambang:.3f}, dihubungi={pred_kapasitas.sum()}, "
      f"churn tertangkap={((pred_kapasitas == 1) & (y_test == 1)).sum()} dari {y_test.sum()}")

imp = permutation_importance(pipe, X_test, y_test, scoring="f1", n_repeats=10, random_state=SEED)
penting = pd.Series(imp.importances_mean, index=X_test.columns).sort_values()
penting.plot.barh(title="Permutation importance (penurunan F1)"); plt.show()

from sklearn.metrics import accuracy_score, precision_score, recall_score, roc_auc_score

metrik = {"accuracy": accuracy_score(y_test, pred), "precision": precision_score(y_test, pred),
          "recall": recall_score(y_test, pred), "f1": f1_score(y_test, pred),
          "roc_auc": roc_auc_score(y_test, proba)}
top = penting.sort_values(ascending=False)
laporan = "\n".join(
    [f"# Laporan Model Churn — {terbaik}", "", "## Metrik Evaluasi", "", "| Metrik | Nilai |", "|---|---|"]
    + [f"| {k} | {v:.3f} |" for k, v in metrik.items()]
    + ["", "## Fitur Terpenting", ""]
    + [f"{i}. {n} ({v:.3f})" for i, (n, v) in enumerate(top.head(3).items(), 1)]
)
print(laporan)

# 2.

# 3.

# 5.