search

brendanhasz / probflow

A Python package for building Bayesian models with TensorFlow or PyTorch
http://probflow.readthedocs.io
MIT License
171 stars 17 forks source link

NotImplementedError:... issue when try to run SimpleLinearRegression example #66

Closed caprone closed 3 years ago

caprone commented 3 years ago

HI "NotImplementedError: Cannot convert a symbolic Tensor (gradients/stateless_random_gamma/StatelessRandomGammaV2_grad/sub:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported"

this issue rises when try to fit SimpleLinearRegression model...(the same example on doc ---> a simple 'copy and paste')

brendanhasz commented 3 years ago

Hi! Thanks for posting the issue! I'm assuming you're using the Tensorflow backend? Sounds like this might be a problem with Tensorflow Probability not playing nice with newer Python versions (see https://github.com/tensorflow/probability/issues/1226).

Which Python version are you running, and which versions of tensorflow and tesnsorflow-probability? Running the "Simple Linear Regression" example (I assume you mean this one?) works with Python 3.7 and

tensorflow-probability==0.12.1 or 0.13.0
tensorflow==2.5.0
probflow==2.4.1

See this colab for a working version.

So unfortunately you may have to downgrade your Python (to 3.8 or 3.7), and/or change tensorflow-probability versions (both 0.12.1 and 0.13.0 worked for me with Python 3.7).

caprone commented 3 years ago

HI and thanks yes problem is Python 3.8