Reading a Bias-Variance Diagnosis

A model at Halewood Analytics comes to review with two error rates and the level of error the desk is willing to live with. Error is the fraction of rows the model got wrong, so lower is better. One rate is measured on the rows the model learned from; the other on a validation set it never saw. The review has three possible findings: the model is too simple to have caught the pattern at all, the model has clung too closely to its training rows, or the model is fine as it stands.

Task: Print the review's finding on this model.

Input

One line holding three numbers separated by single spaces: the training error, then the validation error, then the acceptable threshold.

Output

One line, exactly one of high bias, high variance, or good fit. The finding is high bias whenever the training error is strictly above the threshold: a model that cannot get right the rows it learned from has not caught the pattern, and a good score on the validation set does not change that. It is high variance when the training error is at or below the threshold while the validation error is strictly above it. It is good fit only when neither error is above the threshold. An error sitting exactly on the threshold is acceptable.

Example:

Input:
0.3 0.32 0.1

Output:
high bias