Default Type in pm.Mixture

MFritz · April 26, 2018, 11:46am

Hey there,
I am new to pymc3 and tried to build a certain mixture model. Now my question is whether it is intended that
pm.Mixture defaults to using np.float64 for dtype if dtype argument is not set to float32 (Line 82 in https://github.com/pymc-devs/pymc3/blob/master/pymc3/distributions/mixture.py). This caused a type error in theano for my model, while with dtype set it works just fine. As far as I can tell, this is not documented anywhere. As the general guideline seems to be to pass all values as float32 it might be useful to fix this.

Thank you in advance.

junpenglao · April 26, 2018, 2:19pm

You are right - I am not sure why it was implemented this way as well, a better solution should be using the pm.theanof.floatX. Pull request welcome to change this

github.com

pymc-devs/pymc3/blob/master/pymc3/theanof.py#L59


----------
    a : theano variable


Returns
-------
    r : list of tensor variables that are continuous inputs
"""
return typefilter(inputvars(f), continuous_types)




def floatX(X):
"""
Convert a theano tensor or numpy array to theano.config.floatX type.
"""
try:
    return X.astype(theano.config.floatX)
except AttributeError:
    # Scalar passed
    return np.asarray(X, dtype=theano.config.floatX)

Topic		Replies	Views
Cannot create Dirichlet with "float32" type Questions	4	1200	May 17, 2019
Usage of "pm.Data" leads to: "TypeError: expected type_num 5 (NPY_INT32) got 7" Questions	0	405	March 2, 2021
Scipy functions in pymc3 theano	0	892	February 25, 2020
Value error of pm.model Questions	8	597	May 28, 2019
Test_value shape errors with theano.shared Questions	13	2550	March 27, 2018

Default Type in pm.Mixture

Related topics