CODEWITHSUNDEEP
pythonpymc3
Peadar Coyle
Peadar Coyle

Reputation: 2243

Setting deterministic distributions in PyMC3

I'm new to PyMC3 Here is some PyMC2 - do I need to do something specific like compile in Theano to convert this to PyMC3 code? 

 @pymc.deterministic
    def away_theta(home_team=home_team, 
                   away_team=away_team, 
                   home=home, 
                   atts=atts, 
                   defs=defs, 
                   intercept=intercept): 
        return np.exp(intercept + 
                      atts[away_team] + 
                      defs[home_team])

I get an error like

AsTensorError: ('Cannot convert home_theta to TensorType', <class 'pymc.PyMCObjects.Deterministic'>)

Upvotes: 2

Views: 1093

Answers (1)

twiecki
twiecki

Reputation: 1316

Yes, determinstic transformations need to be theano expressions in pymc3. So instead of using np.exp you'd use theano.tensor.exp:

import theano.tensor as T
import pymc3 as pm

with Model():
    ...
    regression = pm.Deterministic('regression', T.exp(intercept + atts[away_team] + defs[home_team]))
    ...

Upvotes: 3

Related Questions