train_model.py

import pandas as pd
import torch
from transformers import (
    AutoTokenizer,
    AutoModelForSequenceClassification,
    Trainer,
    TrainingArguments,
)
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import numpy as np
from datasets import Dataset, DatasetDict

# Load dataset
df = pd.read_csv("ml/data/request_data.csv")

# Fill missing comments
df["comments"] = df["comments"].fillna("")

# Combine into single text field
df["text"] = (
    df["category"].fillna("") + " - " +
    df["subcategory"].fillna("") + " in " +
    df["area"].fillna("") + ". " +
    df["comments"].fillna("")
)

# Ensure numeric priority labels (0–5)
df["priority"] = pd.to_numeric(df["priority"], errors="coerce").fillna(0).astype(int)
df = df[["text", "priority"]]

# Split train/test
train_df, test_df = train_test_split(df, test_size=0.2, random_state=42)
dataset = DatasetDict({
    "train": Dataset.from_pandas(train_df),
    "test": Dataset.from_pandas(test_df)
})

# 🔁 USE MULTILINGUAL TOKENIZER
model_name = "distilbert-base-multilingual-cased"
tokenizer = AutoTokenizer.from_pretrained(model_name)

def tokenize(batch):
    tokenized = tokenizer(batch["text"], padding=True, truncation=True)
    tokenized["labels"] = batch["priority"]
    return tokenized

dataset = dataset.map(tokenize, batched=True)

# 🔁 USE MULTILINGUAL MODEL
model = AutoModelForSequenceClassification.from_pretrained(model_name, num_labels=6)

# Metric
def compute_metrics(eval_pred):
    logits, labels = eval_pred
    preds = np.argmax(logits, axis=1)
    return {"accuracy": accuracy_score(labels, preds)}

# Training arguments
training_args = TrainingArguments(
    output_dir="./ml/model",
    evaluation_strategy="epoch",
    logging_dir="./ml/model/logs",
    logging_steps=10,
    load_best_model_at_end=True,
    metric_for_best_model="accuracy",
    greater_is_better=True,
    per_device_train_batch_size=8,
    per_device_eval_batch_size=8,
    num_train_epochs=3,
    save_total_limit=2,
)

# Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=dataset["train"],
    eval_dataset=dataset["test"],
    tokenizer=tokenizer,
    compute_metrics=compute_metrics
)

# Train
trainer.train()

# Save model + tokenizer
model.save_pretrained("./ml/model")
tokenizer.save_pretrained("./ml/model/tokenizer")

print("✅ Multilingual model trained and saved to ./ml/model/")