Occasionally, NaN is returned even though the target anchor was evaluated.

[fmohr]

It happens for example if the train curve is ignored.

ipl = lambda x: -0.2 * x **-0.4 + 0.3
evaluator_anomal_learning_curves = lambda learner, anchor, timeout: (ipl(anchor) - 0.01, ipl(anchor)) # train curve is worse than validation curve
score, score_est, elc, model = lccv(sklearn.tree.DecisionTreeClassifier(), None, None, r = 0.29, evaluator = evaluator_anomal_learning_curves, target_anchor = 1500, use_train_curve = False)
print(f"Initial training performance: {model.df.query('trainsize == 64')['score_train'].mean()}")
print(score, elc) ## without this flag, the curve is computed completely