tasks/text.py

from fastapi import APIRouter
from datetime import datetime
from datasets import load_dataset
from sklearn.metrics import accuracy_score
import random

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

router = APIRouter()

DESCRIPTION = "First 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: Random Baseline
    - Makes random predictions from the label space (0-7)
    - Used as a baseline for comparison
    """
    # 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"]
    test_dataset = dataset["test"]
    
    # Start tracking emissions
    tracker.start()
    tracker.start_task("inference")

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE CODE HERE
    # Update the code below to replace the random baseline by your model inference within the inference pass where the energy consumption and emissions are tracked.
    #--------------------------------------------------------------------------------------------   
    class CovidTwitterBertClassifier(nn.Module):
    
        def __init__(self, n_classes):
            super().__init__()
            self.n_classes = n_classes
            self.bert = BertForPreTraining.from_pretrained('digitalepidemiologylab/covid-twitter-bert-v2')    
            self.bert.cls.seq_relationship = nn.Linear(1024, n_classes)
    
            self.sigmoid = nn.Sigmoid()
            
        def forward(self, input_ids, token_type_ids, input_mask):
            outputs = self.bert(input_ids = input_ids, token_type_ids = token_type_ids, attention_mask = input_mask)
    
            logits = outputs[1]
            
            return logits  

    model = CovidTwitterBertClassifier(8)
    
    model.to(device)
    model.load_state_dict(torch.load('model.pth'))
    model.eval()


    tokenizer = AutoTokenizer.from_pretrained('digitalepidemiologylab/covid-twitter-bert')
    
    test_texts = [t['quote'] for t in data_test]

    MAX_LEN = 128 #1024 # < m some tweets will be truncated
    
    tokenized_test = tokenizer(test_texts, max_length=MAX_LEN, padding='max_length', truncation=True)
    test_input_ids, test_token_type_ids, test_attention_mask = tokenized_test['input_ids'], tokenized_test['token_type_ids'], tokenized_test['attention_mask']
    test_token_type_ids = torch.tensor(test_token_type_ids)
    
    test_input_ids = torch.tensor(test_input_ids)
    test_attention_mask = torch.tensor(test_attention_mask)

    batch_size = 8 #
    test_data = TensorDataset(test_input_ids, test_attention_mask, test_token_type_ids)
    
    test_sampler = SequentialSampler(test_data)
    test_dataloader = DataLoader(test_data, sampler=test_sampler, batch_size=batch_size)

    predictions = []
    for step, batch in enumerate(test_dataloader):

        # Add batch to GPU
        batch = tuple(t.to(device) for t in batch)

        b_input_ids, b_input_mask, b_token_type_ids = batch
        with torch.no_grad():
            logits = model(b_input_ids, b_token_type_ids, b_input_mask)    

        logits = logits.detach().cpu().numpy()
        predictions.extend(logits.argmax(1))
        for l in ground_truth:
            labels_sep.append(l)

    
    true_labels = test_dataset["label"]   
    # Make random predictions (placeholder for actual model inference)
    #true_labels = test_dataset["label"]
    #predictions = [random.randint(0, 7) for _ in range(len(true_labels))]

    #--------------------------------------------------------------------------------------------
    # YOUR MODEL INFERENCE STOPS HERE
    #--------------------------------------------------------------------------------------------   

    
    # 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