simon_o
September 24, 2018, 1:44pm
4
There are many things that could be going wrong here, but maybe as a starting point you could have a look at the ADVI troubleshooting/tuning tips found in the thread:
I tried training a linear regression model using both ADVI and MCMC. ADVI is expected to have less accuracy than MCMC when both models are converged. However, I found that ADVI has unacceptable accuracy compare to the MCMC sampling.
The script that was used for evaluation :
import time
import numpy as np
import pymc3 as pm
from sklearn.datasets import load_diabetes, load_boston
from sklearn.metrics import mean_squared_error
from sklearn.model_selection import train_test_split
from theano impo…