Parallel sampling for smc causes attribute error

[jjruby09]

When running sample_smc(parellel=True) on any models AttributeError: 'ValidatingScratchpad' object has no attribute 'test_value' is thrown

Please provide a minimal, self-contained, and reproducible example.

import pymc3 as pm
import numpy as np

data = np.random.randn(500)
with pm.Model() as model:
    mu = pm.Normal("mu", mu=0, sigma=1)
    sd = pm.HalfNormal("sd", sigma=1)
    obs = pm.Normal("obs", mu=mu, sigma=sd, observed=data)

    trace= pm.sample_smc(parallel=True)

Please provide the full traceback.

---------------------------------------------------------------------------
RemoteTraceback                           Traceback (most recent call last)
RemoteTraceback: 
"""
Traceback (most recent call last):
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/multiprocessing/pool.py", line 125, in worker
    result = (True, func(*args, **kwds))
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/multiprocessing/pool.py", line 51, in starmapstar
    return list(itertools.starmap(args[0], args[1]))
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/site-packages/pymc3/smc/sample_smc.py", line 262, in sample_smc_int
    smc.setup_kernel()
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/site-packages/pymc3/smc/smc.py", line 129, in setup_kernel
    self.prior_logp_func = logp_forw([self.model.varlogpt], self.variables, shared)
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/site-packages/pymc3/smc/smc.py", line 278, in logp_forw
    out_list, inarray0 = join_nonshared_inputs(out_vars, vars, shared)
  File "/Users/jruby/anaconda3/envs/pymc3_10_test/lib/python3.8/site-packages/pymc3/theanof.py", line 269, in join_nonshared_inputs
    inarray.tag.test_value = joined.tag.test_value
AttributeError: 'ValidatingScratchpad' object has no attribute 'test_value'
"""

The above exception was the direct cause of the following exception:

AttributeError                            Traceback (most recent call last)
<ipython-input-51-d9c09c21d627> in <module>
      8     obs = pm.Normal("obs", mu=mu, sigma=sd, observed=data)
      9 
---> 10     trace= pm.sample_smc(parallel=True)

~/anaconda3/envs/pymc3_10_test/lib/python3.8/site-packages/pymc3/smc/sample_smc.py in sample_smc(draws, kernel, n_steps, start, tune_steps, p_acc_rate, threshold, save_sim_data, save_log_pseudolikelihood, model, random_seed, parallel, chains, cores)
    189         loggers = [_log] + [None] * (chains - 1)
    190         pool = mp.Pool(cores)
--> 191         results = pool.starmap(
    192             sample_smc_int, [(*params, random_seed[i], i, loggers[i]) for i in range(chains)]
    193         )

~/anaconda3/envs/pymc3_10_test/lib/python3.8/multiprocessing/pool.py in starmap(self, func, iterable, chunksize)
    370         `func` and (a, b) becomes func(a, b).
    371         '''
--> 372         return self._map_async(func, iterable, starmapstar, chunksize).get()
    373 
    374     def starmap_async(self, func, iterable, chunksize=None, callback=None,

~/anaconda3/envs/pymc3_10_test/lib/python3.8/multiprocessing/pool.py in get(self, timeout)
    769             return self._value
    770         else:
--> 771             raise self._value
    772 
    773     def _set(self, i, obj):

AttributeError: 'ValidatingScratchpad' object has no attribute 'test_value'

sample() and sample_smc(parallel=False) successfully execute with proper results. I have tried to messing with theano.config.compute_test_value = X according to #562 but have not found a workable solution. All other functionality I have tried has been succesful thus far including saving and loading Arviz inference objects and testing a number of Arviz plots (using the other sampling techniques).

Versions and main components

• PyMC3 Version: 3.10.0
• Theano Version: (Theano-PYMC) 1.0.11
• Python Version: 3.8
• Operating system: OSX 10.14.6
• How did you install PyMC3: pip within anaconda (conda installs were unsuccessful)

[gracebato]

Hi,

I have the same problem when using smc sampling. Any fix on running parallel=True ?
Thank you.

[ricardoV94]

This should be fixed in V4.0.0. Closing for the time being