Badly fitting DPMM

I created a toy dataset with bimodal distribution, and am attempting to apply the same code from the DPMM density estimation example.

The problem is that the estimated components don’t fit the distribution well at all. It seems that the individual components never have a small enough SD to properly match with the bimodal distribution.


Do I need to adjust some parameter to allow the mixture components to take on smaller SDs?

Aaaaaaand I realized I was accidentally inputting Tau to the SD param of normal mixture…