for n_epoch in range(args.nb_epochs):
log_string(f"--- epoch {n_epoch} ----------------------------------------")
+ cta = " ".join([f"{float(m.main_test_accuracy):.04f}" for m in models])
+ log_string(f"current_test_accuracies {cta}")
+
# Select, improve, and eval the worst model
weakest_model = min(models, key=lambda m: float(m.main_test_accuracy))
f"test_set_composition w_quizzes {quizz_machine.nb_batch_w_quizzes} c_quizzes {quizz_machine.nb_batch_c_quizzes}"
)
- cta = " ".join([f"{float(m.main_test_accuracy):.04f}" for m in models])
- log_string(f"current_test_accuracies {cta}")
-
# Replace a fraction of the w_quizzes with fresh ones
quizz_machine.renew_w_quizzes(args.nb_train_samples // args.nb_gpts)