tasks/text.py

from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
from transformers import DistilBertTokenizerFast, DistilBertForSequenceClassification
from transformers import Trainer, TrainingArguments
import torch

from .utils.evaluation import TextEvaluationRequest
from .utils.emissions import tracker, clean_emissions_data, get_space_info

router = APIRouter()

DESCRIPTION = "DistilBERT Baseline"
ROUTE = "/text"

@router.post(ROUTE, tags=["Text Task"], 
             description=DESCRIPTION)
async def evaluate_text(request: TextEvaluationRequest):
    """
    Evaluate text classification for climate disinformation detection.
    
    Current Model: DistilBERT
    - Fine-tunes and evaluates a DistilBERT model on the given dataset
    """
    # Get space info
    username, space_url = get_space_info()

    # Define the label mapping
    LABEL_MAPPING = {
        "0_not_relevant": 0,
        "1_not_happening": 1,
        "2_not_human": 2,
        "3_not_bad": 3,
        "4_solutions_harmful_unnecessary": 4,
        "5_science_unreliable": 5,
        "6_proponents_biased": 6,
        "7_fossil_fuels_needed": 7
    }

    # Load and prepare the dataset
    dataset = load_dataset(request.dataset_name)

    # Convert string labels to integers
    dataset = dataset.map(lambda x: {"label": LABEL_MAPPING[x["label"]]})

    # Split dataset
    train_test = dataset["train"].train_test_split(test_size=request.test_size, seed=request.test_seed)
    train_dataset = train_test["train"]
    test_dataset = train_test["test"]

    # Tokenizer and model
    tokenizer = DistilBertTokenizerFast.from_pretrained("distilbert-base-uncased")
    model = DistilBertForSequenceClassification.from_pretrained("distilbert-base-uncased", num_labels=8)

    # Tokenize datasets
    def preprocess(examples):
        return tokenizer(examples["text"], truncation=True, padding=True, max_length=512)

    train_dataset = train_dataset.map(preprocess, batched=True)
    test_dataset = test_dataset.map(preprocess, batched=True)

    train_dataset.set_format(type="torch", columns=["input_ids", "attention_mask", "label"])
    test_dataset.set_format(type="torch", columns=["input_ids", "attention_mask", "label"])

    # Training arguments
    training_args = TrainingArguments(
        output_dir="./results",
        evaluation_strategy="epoch",
        learning_rate=5e-5,
        per_device_train_batch_size=16,
        per_device_eval_batch_size=16,
        num_train_epochs=3,
        weight_decay=0.01,
        logging_dir="./logs",
        logging_steps=10,
    )

    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
        eval_dataset=test_dataset,
        tokenizer=tokenizer,
    )

    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    # Train and evaluate the model
    trainer.train()

    # Perform inference
    predictions = trainer.predict(test_dataset).predictions
    predictions = torch.argmax(torch.tensor(predictions), axis=1).tolist()
    true_labels = test_dataset["label"]

    # Stop tracking emissions
    emissions_data = tracker.stop_task()

    # Calculate accuracy
    accuracy = accuracy_score(true_labels, predictions)

    # Prepare results dictionary
    results = {
        "username": username,
        "space_url": space_url,
        "submission_timestamp": datetime.now().isoformat(),
        "model_description": DESCRIPTION,
        "accuracy": float(accuracy),
        "energy_consumed_wh": emissions_data.energy_consumed * 1000,
        "emissions_gco2eq": emissions_data.emissions * 1000,
        "emissions_data": clean_emissions_data(emissions_data),
        "api_route": ROUTE,
        "dataset_config": {
            "dataset_name": request.dataset_name,
            "test_size": request.test_size,
            "test_seed": request.test_seed
        }
    }
    
    return results