Open rgrohit opened 9 years ago
I was trying to figure this out as well.
I _think_ how it works is that your data is providing you a distribution and since your hypothesis is defining parameters you can then find the likelihood of seeing your data under your model with the hypothesis parameters. This corresponds to the risk of accepting/rejecting the null hypothesis given the corresponding parameters.
I found this to be pretty useful. http://math.arizona.edu/~jwatkins/hypothesis.pdf
In the action space we said we reject or fail to reject our hypothesis. In order to this, do we need an expected distribution to test our data against? How do you find the distribution to use?