import os
os.environ["TF_USE_LEGACY_KERAS"] = "1"  # Bypass Keras 3 & optree

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

from datasets import load_dataset

from sklearn.model_selection import train_test_split
from sklearn.metrics import (
    classification_report,
    confusion_matrix,
    ConfusionMatrixDisplay
)

import tensorflow as tf
import tf_keras as keras
from tf_keras import layers
from tf_keras.callbacks import EarlyStopping
from tf_keras.preprocessing.text import Tokenizer  # type: ignore
from tf_keras.preprocessing.sequence import pad_sequences

# ============================================================
# KONFIGURASI DIREKTORI PENYIMPANAN
# ============================================================

VOCAB_SIZE = 10000
MAX_LEN = 60
EMBED_DIM = 64

EPOCHS = 20
BATCH_SIZE = 64
RANDOM_STATE = 42

FIGURES_DIR = "results/figures"
REPORTS_DIR = "results/reports"
MODELS_DIR = "results/models"

os.makedirs(FIGURES_DIR, exist_ok=True)
os.makedirs(REPORTS_DIR, exist_ok=True)
os.makedirs(MODELS_DIR, exist_ok=True)


# ============================================================
# SEL 7 — LOAD DATASET TOKOPEDIA
# ============================================================

print("=" * 60)
print("SEL 7 — LOAD DATASET")
print("=" * 60)

dataset = load_dataset(
    "farhamu/tokopedia-product-reviews-2019"
)

df = dataset["train"].to_pandas()

# Buang rating 3
df = df[df["rating"] != 3].copy()

# Rating 4-5 = positif (1), Rating 1-2 = negatif (0)
df["label"] = (df["rating"] >= 4).astype(int)

# Rename text menjadi teks
df = df.rename(
    columns={"text": "teks"}
)[["teks", "label"]].reset_index(drop=True)

print(df.head(5))
print(f"\nShape : {df.shape}")
print(f"Kolom : {df.columns.tolist()}")
print(f"\nDistribusi label:\n{df['label'].value_counts().sort_index()}")
print(f"\nNull : {df.isnull().sum().sum()}")


# ============================================================
# DISTRIBUSI LABEL
# ============================================================

label_counts = df["label"].value_counts().sort_index()

plt.figure(figsize=(7, 5))
plt.bar(
    ["Negatif", "Positif"],
    [label_counts.get(0, 0), label_counts.get(1, 0)]
)
plt.title("Distribusi Label Sentimen Tokopedia")
plt.xlabel("Sentimen")
plt.ylabel("Jumlah Review")
plt.tight_layout()

plt.savefig(f"{FIGURES_DIR}/distribusi_label_sentimen.png")
plt.close()


# ============================================================
# CEK PANJANG TEKS
# ============================================================

df["panjang"] = df["teks"].apply(lambda x: len(str(x).split()))
print(f"\nPanjang teks:\n{df['panjang'].describe()}")


# ============================================================
# SEL 8 — TEXT PREPROCESSING
# ============================================================

print("\n" + "=" * 60)
print("SEL 8 — TEXT PREPROCESSING")
print("=" * 60)

def clean_text(teks):
    teks = str(teks).lower()
    teks = re.sub(r"[^a-z\s]", "", teks)
    teks = re.sub(r"\s+", " ", teks).strip()
    return teks

df["teks_bersih"] = df["teks"].apply(clean_text)

# Train/Test Split
X = np.array(df["teks_bersih"].tolist())
y = np.array(df["label"].tolist(), dtype=int)

X_tr, X_te, y_tr, y_te = train_test_split(
    X, y,
    test_size=0.2,
    random_state=RANDOM_STATE,
    stratify=y
)

print(f"Training data : {len(X_tr)}")
print(f"Testing data  : {len(X_te)}")

# Tokenizer
tokenizer = Tokenizer(num_words=VOCAB_SIZE, oov_token="<OOV>")
tokenizer.fit_on_texts(X_tr)

# Text -> Integer Sequence
X_tr_seq = tokenizer.texts_to_sequences(X_tr)
X_te_seq = tokenizer.texts_to_sequences(X_te)

# Padding
X_tr_pad = pad_sequences(X_tr_seq, maxlen=MAX_LEN, padding="post")
X_te_pad = pad_sequences(X_te_seq, maxlen=MAX_LEN, padding="post")

print(f"\nX_tr_pad shape : {X_tr_pad.shape}")
print(f"X_te_pad shape : {X_te_pad.shape}")
print(f"Vocab size     : {len(tokenizer.word_index)}")


# ============================================================
# SEL 9 — MODEL GAP
# ============================================================

print("\n" + "=" * 60)
print("SEL 9 — MODEL GAP")
print("=" * 60)

sent_model = keras.Sequential([
    layers.Embedding(input_dim=VOCAB_SIZE, output_dim=EMBED_DIM, input_length=MAX_LEN),
    layers.GlobalAveragePooling1D(),
    layers.Dense(64, activation="relu"),
    layers.Dropout(0.4),
    layers.Dense(1, activation="sigmoid")
], name="Sentimen_GAP")

sent_model.compile(
    optimizer="adam",
    loss="binary_crossentropy",
    metrics=["accuracy"]
)

sent_model.summary()


# ============================================================
# TRAINING
# ============================================================

print("\n" + "=" * 60)
print("TRAINING MODEL GAP")
print("=" * 60)

early_stop = EarlyStopping(
    monitor="val_loss",
    patience=4,
    restore_best_weights=True
)

history_sent = sent_model.fit(
    X_tr_pad, y_tr,
    epochs=EPOCHS,
    batch_size=BATCH_SIZE,
    validation_split=0.15,
    callbacks=[early_stop],
    verbose=1
)


# ============================================================
# TEST ACCURACY & HISTORY PLOTS
# ============================================================

loss_s, acc_s = sent_model.evaluate(X_te_pad, y_te, verbose=0)
print(f"\nSentimen GAP → Test accuracy: {acc_s:.4f}")

# Plot Accuracy & Loss
plt.figure(figsize=(8, 5))
plt.plot(history_sent.history["accuracy"], label="Training Accuracy")
plt.plot(history_sent.history["val_accuracy"], label="Validation Accuracy")
plt.title("Training vs Validation Accuracy — GAP")
plt.xlabel("Epoch")
plt.ylabel("Accuracy")
plt.legend()
plt.tight_layout()
plt.savefig(f"{FIGURES_DIR}/histori_gap_sentimen.png")
plt.close()


# ============================================================
# SEL 11 — EVALUATION
# ============================================================

print("\n" + "=" * 60)
print("EVALUATION")
print("=" * 60)

y_pred_prob = sent_model.predict(X_te_pad, verbose=0)
y_pred = (y_pred_prob > 0.5).astype("int").flatten()

report = classification_report(
    y_te, y_pred,
    target_names=["Negatif (0)", "Positif (1)"]
)

print("\nClassification Report:")
print(report)

# Simpan Classification Report Txt
with open(f"{REPORTS_DIR}/classification_report_gap_sentimen.txt", "w", encoding="utf-8") as f:
    f.write(report)

# Confusion Matrix Plot
cm = confusion_matrix(y_te, y_pred)
disp = ConfusionMatrixDisplay(confusion_matrix=cm, display_labels=["Negatif", "Positif"])
fig, ax = plt.subplots(figsize=(6, 6))
disp.plot(ax=ax, cmap="Blues")
plt.title("Confusion Matrix — Model Sentimen GAP")
plt.tight_layout()
plt.savefig(f"{FIGURES_DIR}/confusion_matrix_gap_sentimen.png")
plt.close()


# ============================================================
# SEL 12 — INFERENCE (PREDIKSI 5 ULASAN)
# ============================================================

def prediksi_sentimen(teks, model, tokenizer, maxlen):
    teks_bersih = clean_text(teks)
    seq = tokenizer.texts_to_sequences([teks_bersih])
    padded = pad_sequences(seq, maxlen=maxlen, padding="post")
    prob = model.predict(padded, verbose=0)[0][0]
    label = "Positif" if prob > 0.5 else "Negatif"
    return label, prob

print("\n" + "=" * 60)
print("PREDIKSI 5 ULASAN")
print("=" * 60)

ulasan_uji = [
    "Produk berkualitas tinggi, pengiriman cepat, packing aman!",
    "Kecewa banget, barang datang rusak dan customer service tidak responsif",
    "Lumayan sih, tapi harganya kemahalan untuk kualitas segitu",
    "Tidak sesuai deskripsi sama sekali, mau return tapi ribet",
    "Barang bagus, sesuai dengan foto dan penjual sangat ramah"
]

for i, ulasan in enumerate(ulasan_uji, start=1):
    label, prob = prediksi_sentimen(ulasan, sent_model, tokenizer, MAX_LEN)
    print(f"\n{i}. {ulasan}\n   Prediksi : {label}\n   Confidence: {prob:.3f}")


# ============================================================
# SIMPAN MODEL
# ============================================================

sent_model.save(f"{MODELS_DIR}/model_gap_sentimen.keras")
print("\nModel GAP berhasil disimpan di folder results/models/")