app.py

from transformers import RobertaForSequenceClassification, RobertaTokenizer, Trainer, TrainingArguments
from datasets import Dataset, DatasetDict
import pandas as pd
from sklearn.preprocessing import LabelEncoder

# Load the dataset
df = pd.read_csv("processed_step3.csv")

# Prepare the dataset for Hugging Face
def preprocess_data(row):
    return {"text": row["full_text"], "labels": row["narratives"]}

# Apply label encoding to narratives to turn them into numeric labels
label_encoder = LabelEncoder()
df["labels"] = label_encoder.fit_transform(df["narratives"])

# Create a Dataset object
hf_dataset = Dataset.from_pandas(df)

# Split the dataset into train and validation sets (80/20 split)
hf_dataset = hf_dataset.train_test_split(test_size=0.2)

# Load pre-trained tokenizer and model
tokenizer = RobertaTokenizer.from_pretrained("roberta-base")
model = RobertaForSequenceClassification.from_pretrained(
    "roberta-base", num_labels=len(label_encoder.classes_))  # Use the number of unique labels

# Tokenize the data
def tokenize_function(examples):
    return tokenizer(examples["text"], padding="max_length", truncation=True)

hf_dataset = hf_dataset.map(tokenize_function, batched=True)

# Set Hugging Face TrainingArguments
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    save_strategy="epoch",
    per_device_train_batch_size=8,
    num_train_epochs=3,
    load_best_model_at_end=True,
    logging_dir="./logs",
    logging_steps=10,
    push_to_hub=True,  # Push to Hugging Face Model Hub
    hub_model_id="eerrffuunn/semeval-task"
)

# Trainer for training the model
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=hf_dataset["train"],  # Train set
    eval_dataset=hf_dataset["test"],  # Validation set
    tokenizer=tokenizer
)

# Train the model
trainer.train()

# Save the model and tokenizer
trainer.save_model("semeval_model")
tokenizer.save_pretrained("semeval_model")