WARNING (theano.tensor.blas): Using NumPy C-API based implementation for BLAS functions.-------->HELP!

#1

Hello,
I am trying to run a pymc3 script for parameter estimation. It is giving me these warnings:
WARNING (theano.tensor.blas): Using NumPy C-API based implementation for BLAS functions.
WARNING (theano.gof.compilelock): Overriding existing lock by dead process ‘9269’ (I am process ‘11498’)
I am using python ---- 3.7
pymc3----3.6
theano ---- 1.0.3 versions.
I am fairly new to coding and need some help to resolve this.
Any help is much appreciated.
Thank you.

#2

Hi @Nandita_Khetan. The first warning is telling you that theano will run slower because it did not find some libraries installed that help it go faster. You can get these in different ways depending on your operating system and on how you installed theano and pymc3. Are you using windows, mac or another unix operating system? Did you install pymc3 using conda or pip?

The second warning is nothing to be alarmed about. The model’s logp in pymc3 gets compiled by theano. This happens a few times while preparing for sampling. This compilation happens mostly in parallel but there’s a point at which a single process is allowed to be running the compilation (compilelock). When that happens, the other processes have to wait in a queue for the others to finish, and print out the warning you see.

#3

HI. Thanks for help.
I have mac OS Mojave. I installed Pymc3 using conda. I updated my theano and i got rid of the first warning. I restarted my PC and got rid of the second one. However now I am getting this while trying to run the code:
AttributeError: module ‘theano’ has no attribute ‘gof’
:(…any insight??

#4

You broke your theano installation with the update. I suggest you start out in a new conda env, or uninstall everything from the env you were working with and run the following, which I looked up here:

conda install numpy scipy mkl
conda install theano pygpu
conda install pymc3