import time
import numpy as np
import pandas as pd
from tqdm import tqdm  # ✅ Ajout ici
from datasets import load_dataset
from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score, log_loss
from inference import (
    zero_shot_inference,
    few_shot_inference,
    base_model_inference,
)

# Dictionnaire des fonctions à évaluer
models_to_evaluate = {
    "Base model": base_model_inference,
    "Zero-shot": zero_shot_inference,
    "Few-shot": few_shot_inference,
}

label_map = {0: "World", 1: "Sports", 2: "Business", 3: "Sci/Tech"}

# Charger un sous-ensemble du jeu de test AG News
dataset = load_dataset("ag_news", split="test[:10%]")  

def evaluate_model(name, inference_func):
    print(f"\n🔍 Évaluation du modèle : {name}")
    true_labels = []
    pred_labels = []
    all_probs = []

    start = time.time()

    for example in tqdm(dataset, desc=f"Modèle : {name}"):  
        text = example["text"]
        true_label = example["label"]
        true_class = label_map[true_label]

        try:
            pred_class, scores = inference_func(text)
        except Exception as e:
            print(f"⚠️ Erreur sur un exemple : {e}")
            continue

        # Scores pour les 4 classes dans le même ordre
        prob_dist = [scores.get(c, 0.0) for c in label_map.values()]
        pred_index = list(label_map.values()).index(pred_class)

        pred_labels.append(pred_index)
        true_labels.append(true_label)
        all_probs.append(prob_dist)

    end = time.time()
    runtime = round(end - start, 2)

    acc = accuracy_score(true_labels, pred_labels)
    f1 = f1_score(true_labels, pred_labels, average='weighted')
    prec = precision_score(true_labels, pred_labels, average='weighted')
    rec = recall_score(true_labels, pred_labels, average='weighted')
    loss = log_loss(true_labels, all_probs)

    print(f"✅ Résultats {name} :")
    print(f"- Accuracy : {acc:.4f}")
    print(f"- F1 Score : {f1:.4f}")
    print(f"- Precision : {prec:.4f}")
    print(f"- Recall : {rec:.4f}")
    print(f"- Log Loss : {loss:.4f}")
    print(f"- Runtime : {runtime:.2f} sec\n")

    return {
        "model": name,
        "accuracy": acc,
        "f1_score": f1,
        "precision": prec,
        "recall": rec,
        "loss": loss,
        "runtime": runtime
    }

# Évaluer tous les modèles
results = []
for name, func in models_to_evaluate.items():
    results.append(evaluate_model(name, func))

# Affichage résumé
df = pd.DataFrame(results)
print(df)