import gradio as gr

# Function to execute evaluate_model.py
def evaluate_model_script():
    import spacy
    import jsonlines
    from sklearn.metrics import classification_report, accuracy_score, f1_score, precision_score, recall_score

    # Load the trained spaCy model
    nlp = spacy.load("./my_trained_model")

    # Load the golden evaluation data
    golden_eval_data = []
    with jsonlines.open("data/goldenEval.jsonl") as reader:
        for record in reader:
            golden_eval_data.append(record)

    # Predict labels for each record using your model
    predicted_labels = []
    for record in golden_eval_data:
        text = record["text"]
        doc = nlp(text)
        predicted_labels.append(doc.cats)

    # Extract ground truth labels from the golden evaluation data
    true_labels = [record["accept"] for record in golden_eval_data]

    # Convert label format to match sklearn's classification report format
    true_labels_flat = [label[0] if label else "reject" for label in true_labels]
    predicted_labels_flat = [max(pred, key=pred.get) for pred in predicted_labels]

    # Calculate evaluation metrics
    accuracy = accuracy_score(true_labels_flat, predicted_labels_flat)
    precision = precision_score(true_labels_flat, predicted_labels_flat, average='weighted')
    recall = recall_score(true_labels_flat, predicted_labels_flat, average='weighted')
    f1 = f1_score(true_labels_flat, predicted_labels_flat, average='weighted')

    # Additional classification report
    report = classification_report(true_labels_flat, predicted_labels_flat)

    # Build the result dictionary
    result = {
        "accuracy": accuracy,
        "precision": precision,
        "recall": recall,
        "f1_score": f1,
        "detailed_classification_report": report
    }

    return result

# Gradio Interface
output = gr.outputs.Label(type="json", label="Evaluation Metrics")
iface = gr.Interface(fn=evaluate_model_script, outputs=output, title="Evaluate Model Script")
iface.launch()