eval metrics for bio-reasoning v1.

metrics.py

from sklearn.metrics import (
    average_precision_score, 
    f1_score, 
    roc_auc_score, 
    precision_score, 
    recall_score, 
    accuracy_score
)

def evaluate_bi_cls(y_true, y_pred):
    results = {
        'PRAUC': float(average_precision_score(y_true, y_pred)),
        'F1': float(f1_score(y_true, y_pred)),
        'ROCAUC': float(roc_auc_score(y_true, y_pred)),
        'Precision': float(precision_score(y_true, y_pred)),
        'Recall': float(recall_score(y_true, y_pred)),
        'Accuracy': float(accuracy_score(y_true, y_pred))
    }
    return results
    
def ndcg_at_k(predicted, ground_truth, k=10):
    gt_genes = set(g.split(' ')[0] for g in ground_truth)
    relevance = [1 if g.split(' ')[0] in gt_genes else 0 for g in predicted[:k]]
    ideal = sorted(relevance, reverse=True)

    def dcg(scores):
        return sum(rel / np.log2(i + 2) for i, rel in enumerate(scores))

    ideal_dcg = dcg(ideal)
    return dcg(relevance) / ideal_dcg if ideal_dcg > 0 else 0.0
    
def recall_at_k(predicted, ground_truth, k=10):
    top_k_genes = set(g.split(' ')[0] for g in predicted[:k])
    gt_genes = set(g.split(' ')[0] for g in ground_truth)
    matched = top_k_genes & gt_genes
    return len(matched) / len(gt_genes) if gt_genes else 0.0