# Evaluate

{% embed url="<https://www.youtube.com/watch?v=wP6BkXcB_Xg>" %}
Evaluate - Troubleshooting
{% endembed %}

## Summary

* You want to apply **the bias-variance decomposition** concept here: *Test error = irreducible error + bias + variance + validation overfitting*.
* If the training, validation, and test sets come from different data distributions, then you should use **2 validation sets**: one set sampled from the training distribution, and the other set sampled from the test distribution.