#!pip install -U transformers
from transformers import TrainingArguments

training_args = TrainingArguments(
    output_dir="./results",
    learning_rate=2e-5,
    per_device_train_batch_size=4,
    per_device_eval_batch_size=4,
    num_train_epochs=3,
    weight_decay=0.01,
    logging_dir="./logs",
    logging_steps=10
)

!pip uninstall -y transformers
!pip install -U transformers datasets accelerate
!pip show transformers | grep Version

import os
os.environ["WANDB_DISABLED"] = "true"


# ===============================================
# 1️⃣ CÀI ĐẶT THƯ VIỆN
# ===============================================
!pip install -q transformers datasets torch

# ===============================================
# 2️⃣ TẠO DỮ LIỆU GIẢ LẬP (CSV)
# ===============================================
import pandas as pd

data = {
    "text": [
        "I love this movie, it was fantastic!",
        "This product is terrible and useless.",
        "What a great experience, I will come again!",
        "I hate this item, waste of money.",
        "Absolutely amazing service and food.",
        "Worst app I have ever used.",
        "The phone works perfectly and fast.",
        "It broke after two days, horrible!",
        "Very happy with my purchase.",
        "Not worth the price at all."
    ],
    "label": [1,0,1,0,1,0,1,0,1,0]
}

df = pd.DataFrame(data)
df.to_csv("sentiment_data.csv", index=False)
print("✅ Dữ liệu mẫu đã được tạo:\n")
print(df.head())

# ===============================================
# 3️⃣ TẢI DỮ LIỆU & CHUẨN HÓA
# ===============================================
from datasets import load_dataset

dataset = load_dataset("csv", data_files="sentiment_data.csv")
dataset = dataset["train"].train_test_split(test_size=0.3, seed=42)

train_dataset = dataset["train"]
test_dataset = dataset["test"]

print("\n🔹 Số mẫu train:", len(train_dataset))
print("🔹 Số mẫu test:", len(test_dataset))

# ===============================================
# 4️⃣ TOKENIZATION (CHUYỂN TEXT THÀNH INPUT CHO BERT)
# ===============================================
from transformers import AutoTokenizer

model_name = "bert-base-uncased"
tokenizer = AutoTokenizer.from_pretrained(model_name)

def preprocess_function(examples):
    return tokenizer(
        examples["text"],
        padding="max_length",
        truncation=True,
        max_length=64,
    )

train_tokenized = train_dataset.map(preprocess_function, batched=True)
test_tokenized = test_dataset.map(preprocess_function, batched=True)

# ===============================================
# 5️⃣ CHUẨN BỊ MÔ HÌNH BERT CHO PHÂN LOẠI
# ===============================================
import torch
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer

model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2)

# ===============================================
# 6️⃣ ĐỊNH NGHĨA HÀM ĐÁNH GIÁ
# ===============================================
from sklearn.metrics import accuracy_score, f1_score

def compute_metrics(eval_pred):
    logits, labels = eval_pred
    preds = torch.argmax(torch.tensor(logits), dim=1)
    acc = accuracy_score(labels, preds)
    f1 = f1_score(labels, preds)
    return {"accuracy": acc, "f1": f1}

# ===============================================
# 7️⃣ CẤU HÌNH HUẤN LUYỆN
# ===============================================
training_args = TrainingArguments(
    output_dir="./results",
    learning_rate=2e-5,
    per_device_train_batch_size=4,
    per_device_eval_batch_size=4,
    num_train_epochs=3,
    weight_decay=0.01,
    evaluation_strategy="epoch",
    save_strategy="epoch",
    logging_dir="./logs",
    logging_steps=10
)

# ===============================================
# 8️⃣ HUẤN LUYỆN MÔ HÌNH
# ===============================================
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=train_tokenized,
    eval_dataset=test_tokenized,
    tokenizer=tokenizer,
    compute_metrics=compute_metrics
)

trainer.train()

# ===============================================
# 9️⃣ ĐÁNH GIÁ
# ===============================================
eval_results = trainer.evaluate()
print("\n📊 Kết quả đánh giá:", eval_results)

# ===============================================
# 🔟 THỬ DỰ ĐOÁN
# ===============================================
text_samples = [
    "I really love this product!",
    "This is the worst movie ever."
]

inputs = tokenizer(text_samples, padding=True, truncation=True, max_length=64, return_tensors="pt")
outputs = model(**inputs)
preds = torch.argmax(outputs.logits, dim=1)

for text, label in zip(text_samples, preds):
    print(f"\n🗣️ {text}")
    print("➡️ Dự đoán:", "Tích cực (1)" if label == 1 else "Tiêu cực (0)")