I started reading "Bayesian Inference Experienced in Python". Make a note of what you got stuck in before moving it.
I put in pymc, but now it is pymc4.
If you check the book's github,
https://github.com/CamDavidsonPilon/Probabilistic-Programming-and-Bayesian-Methods-for-Hackers
The scripts for pymc2 and pymc3 are included, so when reading a book, install pymc3 and look at the scripts for pymc3 in git.
The required libraries are pymc3 and the backend theano.
pip install pymc3
I've done a lot, but without putting anything in it, first of all, it's all installed with just this pip.
Basically, installing this pymc3 will install the required theano.
theno is the backend that runs behind pymc3.
After that, put in the required jupyter notebook.
pip install jupyter
With this, the first chapter works. It spits out a warning, but it works, so it's okay.
Next, I also added my favorite spyder.
This is the case when running with spyder.
In the cell in10 in Ch1_Introduction_PyMC3.ipynb of jupyter, if you move it with spyder as it is, an error will occur here.
#When I run it with this, I get an error.
with model:
step = pm.Metropolis()
trace = pm.sample(10000, tune=5000,step=step)
The reason for the error is that the multi-core settings don't seem to work. This is fixed here.
with model:
step = pm.Metropolis()
trace = pm.sample(10000, tune=5000,step=step , cores = 1)
For the time being, cores = 1, and if you move it with a single core, it will work. If you enter 2 or 4 here, an error will occur, but if something works and you force termination, you will get a result. Set 1 to move normally.
It works with jupyter notebook, but I don't know why I get an error with spyder.
From now on, it seems that tensorflow will be the backend in pymc4, but in the case of this book, it is better to include pymc3.
https://yukinagae.hatenablog.com/entry/2018/09/20/093137
Recommended Posts