How do I adjust to these errors in expectation and judgement? I think the answer lies in understanding that all models are wrong and not every variable and parameter can be accounted for. Sometimes your model is correct but the data doesn't fit, other times the data fits but the model is irrational and nonsensical. Understanding these subtle differences and variations will help with adapting the Bayesian brain and optimizing the search.
While this sounds rosy and rational there are curve balls you need to watch out for. Sometimes the data is non-linear to such an extent that there is no fit. There maybe hidden variables unaccounted for or it is the single outlier that throws any rational model off its trajectory. In such cases, terminating the search should be an option. If you cannot optimize, accept defeat and blame the data. This is what any good algorithm would do. Its never the algorithm its always the data!!
No comments:
Post a Comment