AttributeError: Module 'pymc3.gp' has no attribute 'GP'

[MCruzReyes]

I was trying to reproduce the notebook for Gaussian Processes given in
https://pymc3-testing.readthedocs.io/en/rtd-docs/notebooks/GP-introduction.html ,
however when I type the following lines of code:

# covariance functions for the function f and the noise
    f_cov = s2_f * pm.gp.cov.ExpQuad(1, l)

    y_obs = pm.gp.GP('y_obs', cov_func=f_cov, sigma=s2_n, observed={'X':X, 'Y':y})

I get:

**
AttributeError Traceback (most recent call last)
in
16 f_cov = s2_f * pm.gp.cov.ExpQuad(1, l)
17
---> 18 y_obs = pm.gp.GP('y_obs', cov_func=f_cov, sigma=s2_n, observed={'X':X, 'Y':y})

AttributeError: module 'pymc3.gp' has no attribute 'GP'
**

• PyMC3 Version: pymc3==3.11.2
• Aesara/Theano Version :aesara==2.1.2, theano-pymc==1.1.2
• Python Version: 3.9.6
• Operating system: Windows 10
• PyMC3 was installed usign: Conda

Can someone explaining what I am doing wrong?
Thanks in advance.

[michaelosthege]

That sounds very much like a problem with your installation.
👉 You shouldn't have Aesara installed together with PyMC3 <4
👉 Conda installation of PyMC3 on Windows is somewhat brittle. I am on Windows myself and generally recommend to follow https://github.com/pymc-devs/pymc3/wiki/Installation-Guide-(Windows)

However, that also doesn't work for everybody (e.g. #4749 (comment)) and we have not figured out why.

[MCruzReyes]

Now I changed to Linux, but I have exactly the same problem that was describe above:

**
AttributeError Traceback (most recent call last)
in
16 f_cov = s2_f * pm.gp.cov.ExpQuad(1, l)
17
---> 18 y_obs = pm.gp.GP('y_obs', cov_func=f_cov, sigma=s2_n, observed={'X':X, 'Y':y})

AttributeError: module 'pymc3.gp' has no attribute 'GP'
**
Now I am using:

PyMC3 Version: pymc3==3.11.2
Theano Version : theano-pymc==1.1.2 (Without Aesara)
Python Version: 3.9.6
Operating system: Ubuntu 20.04.2 LTS
PyMC3 was installed usign: Conda

Maybe the problem is related with the Python version? Any recomendations?

Note 1: I have the same problem with Python 3.7.10 and 3.8.10

To install pymc3 I did this procedure:

1. I created the conda environment
2. Clone the repository with: git clone https://github.com/pymc-devs/pymc3
3. I installed of requirements.txt
4. I installed of theano-pymc with: conda install -c conda-forge theano-pymc -y
5. I installed of pymc3 with: conda install -c conda-forge pymc3

[michaelosthege]

If you cloned the repo from https://github.com/pymc-devs/pymc3 you have the latest development version which already depends on Aesara.
I just created an environment successfully with these steps:

1. conda create -n pm3v4 -c conda-forge "python=3.8" libpython mkl-service m2w64-toolchain numba python-graphviz scipy
2. conda activate pm3v4
3. git clone https://github.com/pymc-devs/pymc3
4. cd pymc3
5. pip install -c .

That latest version is still missing a few distributions, but for simple things the GP submodule should be fine already.

[MCruzReyes]

I tried in Windows and Linux and I can't follow your instructions, in both operating systems I get:

** ERROR: Could not open requirements file: [Errno 13] Permission denied: '.' **

[michaelosthege]

That sounds like you did something as root. Either cloning or installing your Python env.
Make sure to NEVER install Python/miniconda/anaconda as root/Admin, because it always causes problems down the line.

[bwengals]

Installation issues aside, I think there isn't a pm.gp.GP class, I think it had gotten split into Marginal and Latent. Also, isn't the pymc3 current documentation here https://docs.pymc.io/ and not here https://pymc3-testing.readthedocs.io/en/rtd-docs/ ?

[MCruzReyes]

Thanks @bwengals, this is the solution to my problem.