pymc warning: value is neither numerical nor array with floating-point dtype

Question

I have a Bayes net (DAG) model which I created using pymc 2.3. All the variables in it are Bernoulli random variables. When I call the MAP.fit() method on it before sampling I get the following warning for all of my random variables:

value is neither numerical nor array with floating-point dtype. Recommend fitting method fmin (default)

From the github repo of pymc, it seems that this warning is printed if the underlying type of the stochastic variable is not a float. For Bernoulli RV the type is (and should be) bool.

Does this mean that the MAP step will have unstable results?

Midge

So, did u get the answer ? – Junket

No, I wasn't able to verify it this way or the other. For my project I couldn't see any difference between using the MAP step or not. – Midge

what do you exactly mean by unstable results? or do you rather ask if the results make sense? Are you solely interested in finding good initial values for your sampling to decrease burnin time or are you interested in the maximum a posterior values? – Laural