Denominator values too small in predict_dpi

[mkirshner]

Hello,

I know this code has not been updated in a while, but I made some changes and was able to gather data and alter the millionaire.py file for testing. However, after debugging, I found that when the predict_dpi method in bayesian_regression.py runs, some of the calculated exponential values are so small that even when using the BigFloat library, Python assumes they are 0. Even summing all these up sometimes produces a value of 0, and so then the code divides by 0, creating NaN values. These NaN values prevent the rest of the code from executing as intended. Is there any way to circumvent this issue?