import os
import joblib


# =========================================================
# KONFIGURASI MODEL
# =========================================================

BASE_DIR = os.path.dirname(
    os.path.dirname(
        os.path.abspath(__file__)
    )
)

MODEL_DIR = os.path.join(
    BASE_DIR,
    "models"
)

MODEL_PATH = os.path.join(
    MODEL_DIR,
    "model_naive_bayes.pkl"
)


# =========================================================
# LOAD MODEL
# =========================================================

if not os.path.exists(MODEL_PATH):
    raise FileNotFoundError(
        f"Model tidak ditemukan: {MODEL_PATH}"
    )

try:
    model_final = joblib.load(MODEL_PATH)

except Exception as e:
    raise RuntimeError(
        f"Gagal memuat model: {str(e)}"
    )


# =========================================================
# VALIDASI PIPELINE
# =========================================================

if not hasattr(model_final, "named_steps"):
    raise ValueError(
        "Model harus berupa Pipeline "
        "TF-IDF + MultinomialNB."
    )


# Pastikan step TF-IDF tersedia
if "tfidf" not in model_final.named_steps:
    raise ValueError(
        "Pipeline tidak memiliki step 'tfidf'."
    )


# Pastikan step Naïve Bayes tersedia
if "nb" not in model_final.named_steps:
    raise ValueError(
        "Pipeline tidak memiliki step 'nb'."
    )


# =========================================================
# AMBIL KOMPONEN MODEL
# =========================================================

tfidf_vectorizer = model_final.named_steps["tfidf"]

nb_model = model_final.named_steps["nb"]


# =========================================================
# PREDIKSI SENTIMEN
# =========================================================

def predict_sentiment(text):

    if not text or not str(text).strip():
        raise ValueError(
            "Teks komentar tidak boleh kosong."
        )

    text = str(text).strip()

    # Prediksi
    prediction = model_final.predict([text])[0]

    # Probabilitas
    probabilities = model_final.predict_proba([text])[0]

    # Konversi label
    sentiment = (
        "Positif"
        if int(prediction) == 1
        else "Negatif"
    )

    # Probabilitas kelas yang diprediksi
    probability = float(
        probabilities[
            list(model_final.classes_).index(prediction)
        ]
    )

    return {
        "label": int(prediction),
        "sentiment": sentiment,
        "probability": round(
            probability,
            6
        )
    }


# =========================================================
# TF-IDF SATU KOMENTAR
# =========================================================

def get_tfidf(text):

    if not text or not str(text).strip():
        return []

    text = str(text).strip()

    # Transform menggunakan vectorizer
    # yang SUDAH FIT dari model final
    vector = tfidf_vectorizer.transform(
        [text]
    )

    feature_names = (
        tfidf_vectorizer
        .get_feature_names_out()
    )

    hasil = []

    # Karena sparse matrix,
    # ambil hanya fitur yang nilainya > 0
    row = vector[0]

    for col_index, nilai in zip(
        row.indices,
        row.data
    ):

        hasil.append({
            "kata": feature_names[col_index],
            "nilai": round(
                float(nilai),
                6
            )
        })

    # Urutkan berdasarkan nilai TF-IDF terbesar
    hasil.sort(
        key=lambda x: x["nilai"],
        reverse=True
    )

    return hasil