Pymc3 Installation Issues in Linux

[Hyunglok-Kim]

Hi,
I am trying to install pymc3 using the following command line in Linux:

conda create -c conda-forge -n pymc3_env pymc3 theano-pymc mkl mkl-service

However, I got the following error when I try to import pymc3, and I suspect this is related to the numpy version:

import pymc3
/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/theano/scalar/basic.py:2412: FutureWarning: In the future np.bool will be defined as the corresponding NumPy scalar.
self.ctor = getattr(np, o_type.dtype)
Traceback (most recent call last):
File "", line 1, in
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/pymc3/init.py", line 23, in
import theano
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/theano/init.py", line 83, in
from theano import scalar, tensor
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/theano/scalar/init.py", line 1, in
from .basic import *
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/theano/scalar/basic.py", line 2460, in
convert_to_bool = Cast(bool, name="convert_to_bool")
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/theano/scalar/basic.py", line 2412, in init
self.ctor = getattr(np, o_type.dtype)
File "/home/h/anaconda3/envs/pymc3/lib/python3.10/site-packages/numpy/init.py", line 305, in getattr
raise AttributeError(former_attrs[attr])
AttributeError: module 'numpy' has no attribute 'bool'.
np.bool was a deprecated alias for the builtin bool. To avoid this error in existing code, use bool by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use np.bool_ here.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations. Did you mean: 'bool_'?

Could you please provide me with any advise on this issue?
Thank you.

[twiecki]

Any reason you can't use PyMC 5.x?

[Hyunglok-Kim]

Hi,

Yes, I am attempting to utilize the continuous zero-inflated exponential likelihood distribution using pm.DensityDist with:

def  logp_zero_inflated_exponential(value, psi, mu):
     return tt.switch(tt.eq(value, 0), tt.log(psi), tt.log1p(-psi) - mu * value)

psi = pm.Beta('psi', 1, 1)
z  = pm.HalfNormal('Z', sigma=500)
a  = pm.HalfNormal('α', sigma=1000) # weak informative prior (we know it should be positive) 
b  = pm.Normal('β', mu=0, sigma=1000)

ET_I_est = f_ET_I(SM1, SM2, a, b, dt)
mu = z*np.abs(dsm) - ET_I_est

Y_obs = pm.DensityDist('Y_obs', logp_zero_inflated_exponential, observed={'value': y_obs, 'psi': psi, 'mu': mu})

This is working fine with the version 3, however, when I use it with
Y_obs = pm.DensityDist('Y_obs', logp_zero_inflated_exponential, observed=np.array([y_obs, psi, mu]))
in the version 5, I got the following error:

    860                 Y_obs = pm.DensityDist('Y_obs', logp_zero_inflated_exponential, 
    861                                        observed={'value': y_obs, 'psi': psi, 'mu': mu})
    862             elif pmv == 5:
    863                 Y_obs = pm.DensityDist('Y_obs', logp_zero_inflated_exponential, 
--> 864                                        observed=np.array([y_obs, psi, mu]))
    865 # sampling or fit
    866 rng = 321

ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (3,) + inhomogeneous part.

I might be missing something with the Version 5 API. If you can provide me with instructions on how to use the Version 5 API correctly, that would be really helpful. If it works, I won't need to use pymc3. By the way, the conda-forge pymc3 library installed on macOS works just fine with the mkl-service library. I'm not sure why the conda-forge pymc3 library for Linux comes with a numpy version that is not 1.22.

Thank you for your help.

[twiecki]

@Hyunglok-Kim So I would try and get it to work on PyMC 5.x. As to the question, can you post that on discourse.pymc.io?

[Hyunglok-Kim]

Sure thing. I have posted it on discourse.pymc.io. Check here.

[Hyunglok-Kim]

Additional information:

I use different logp_zero_inflated_exponential function for pymc3 (pmv == 3) and pymc5:

if pmv == 3:
    def logp_zero_inflated_exponential(value, psi, mu):
        return tt.switch(tt.eq(value, 0), tt.log(psi), tt.log1p(-psi) - mu * value)

if pmv == 5:
    def logp_zero_inflated_exponential(value, psi, mu):
      if value == 0:
        return pm.math.log(psi)
      else:
        return pm.math.log1p(-psi) - mu * value