Machine Learning / AI

3534

Aké metriky sú vhodné na hodnotenie modelov s nevyváženými triedami?

Čo je offline RL a aké sú jeho potenciálne úskalia?

# Formulujte a napíšte definíciu toho, čo je roc auc. # Ako sa počíta roc_auc_score? # Napíšte funkciu, ktorá vypočíta roc_auc_score podľa napísanej definície. # A pre ľubovoľný počet tried s daným poradím priorít? # Model funguje nasledovne # y_score[i] >= prah -> y_pred[i] = 1 # y_score[i] < prah -> y_pred[i] = 0 def test_roc_aur_score(): y_true = [0, 0, 0, 1, 0, 0] y_score = [0.2, 0.4, 0.31, 0.6, 0.3, 0.7] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 0.8 y_true = [0, 1, 0, 0, 0, 1] y_score = [0.7, 0.6, 0.31, 0.31, 0.3, 0.2] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 0.375 y_true = [0, 0, 0, 1, 0, 0] y_score = [0.2, 0.4, 0.31, 0.8, 0.3, 0.7] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 1 y_true = [0, 0, 0, 0, 0, 0] y_score = [0.2, 0.4, 0.31, 0.8, 0.3, 0.7] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 0 # Bonus testy pre viac tried y_true = [0, 1, 0, 0.5, 0, 1] y_score = [0.7, 0.6, 0.31, 0.31, 0.3, 0.2] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 0. y_true = [0, 0.1, 0.2, 0.3, 0.4, 0.6] y_score = [0, 0.1, 0.2, 0.3, 0.4, 0.6] res_metric = roc_auc_score(y_true, y_score) assert res_metric == 1

Senior