BrokenPipeError: [Errno 32] Broken pipe

[ghost]

PyMC3 Modeling tips and heuristic

When I try to run the example in https://docs.pymc.io/notebooks/PyMC3_tips_and_heuristic.html by jupyter, I got the error.

the code is

class CAR(distribution.Continuous):
    """
    Conditional Autoregressive (CAR) distribution

    Parameters
    ----------
    a : list of adjacency information
    w : list of weight information
    tau : precision at each location
    """
    def __init__(self, w, a, tau, *args, **kwargs):
        super(CAR, self).__init__(*args, **kwargs)
        self.a = a = tt.as_tensor_variable(a)
        self.w = w = tt.as_tensor_variable(w)
        self.tau = tau*tt.sum(w, axis=1)
        self.mode = 0.

    def get_mu(self, x):

        def weigth_mu(w, a):
            a1 = tt.cast(a, 'int32')
            return tt.sum(w*x[a1])/tt.sum(w)

        mu_w, _ = scan(fn=weigth_mu,
                       sequences=[self.w, self.a])

        return mu_w

    def logp(self, x):
        mu_w = self.get_mu(x)
        tau = self.tau
        return tt.sum(continuous.Normal.dist(mu=mu_w, tau=tau).logp(x))

with pm.Model() as model1:
    # Vague prior on intercept
    beta0 = pm.Normal('beta0', mu=0.0, tau=1.0e-5)
    # Vague prior on covariate effect
    beta1 = pm.Normal('beta1', mu=0.0, tau=1.0e-5)

    # Random effects (hierarchial) prior
    tau_h = pm.Gamma('tau_h', alpha=3.2761, beta=1.81)
    # Spatial clustering prior
    tau_c = pm.Gamma('tau_c', alpha=1.0, beta=1.0)

    # Regional random effects
    theta = pm.Normal('theta', mu=0.0, tau=tau_h, shape=N)
    mu_phi = CAR('mu_phi', w=wmat, a=amat, tau=tau_c, shape=N)

    # Zero-centre phi
    phi = pm.Deterministic('phi', mu_phi-tt.mean(mu_phi))

    # Mean model
    mu = pm.Deterministic('mu', tt.exp(logE + beta0 + beta1*aff + theta + phi))

    # Likelihood
    Yi = pm.Poisson('Yi', mu=mu, observed=O)

    # Marginal SD of heterogeniety effects
    sd_h = pm.Deterministic('sd_h', tt.std(theta))
    # Marginal SD of clustering (spatial) effects
    sd_c = pm.Deterministic('sd_c', tt.std(phi))
    # Proportion sptial variance
    alpha = pm.Deterministic('alpha', sd_c/(sd_h+sd_c))

    trace1 = pm.sample(int(3e3), cores=2, tune=1000, nuts_kwargs={'max_treedepth': 15})

And the error

BrokenPipeError                           Traceback (most recent call last)
<ipython-input-7-088cf3f1b726> in <module>
     30     alpha = pm.Deterministic('alpha', sd_c/(sd_h+sd_c))
     31 
---> 32     trace1 = pm.sample(int(3e3), cores=2, tune=1000, nuts_kwargs={'max_treedepth': 15})

C:\Program\lib\site-packages\pymc3\sampling.py in sample(draws, step, init, n_init, start, trace, chain_idx, chains, cores, tune, nuts_kwargs, step_kwargs, progressbar, model, random_seed, live_plot, discard_tuned_samples, live_plot_kwargs, compute_convergence_checks, use_mmap, **kwargs)
    437             _print_step_hierarchy(step)
    438             try:
--> 439                 trace = _mp_sample(**sample_args)
    440             except pickle.PickleError:
    441                 _log.warning("Could not pickle model, sampling singlethreaded.")

C:\Program\lib\site-packages\pymc3\sampling.py in _mp_sample(draws, tune, step, chains, cores, chain, random_seed, start, progressbar, trace, model, use_mmap, **kwargs)
    984         sampler = ps.ParallelSampler(
    985             draws, tune, chains, cores, random_seed, start, step,
--> 986             chain, progressbar)
    987         try:
    988             try:

C:\Program\lib\site-packages\pymc3\parallel_sampling.py in __init__(self, draws, tune, chains, cores, seeds, start_points, step_method, start_chain_num, progressbar)
    311                 draws, tune, step_method, chain + start_chain_num, seed, start
    312             )
--> 313             for chain, seed, start in zip(range(chains), seeds, start_points)
    314         ]
    315 

C:\Program\lib\site-packages\pymc3\parallel_sampling.py in <listcomp>(.0)
    311                 draws, tune, step_method, chain + start_chain_num, seed, start
    312             )
--> 313             for chain, seed, start in zip(range(chains), seeds, start_points)
    314         ]
    315 

C:\Program\lib\site-packages\pymc3\parallel_sampling.py in __init__(self, draws, tune, step_method, chain, seed, start)
    202         )
    203         # We fork right away, so that the main process can start tqdm threads
--> 204         self._process.start()
    205 
    206     @property

C:\Program\lib\multiprocessing\process.py in start(self)
    110                'daemonic processes are not allowed to have children'
    111         _cleanup()
--> 112         self._popen = self._Popen(self)
    113         self._sentinel = self._popen.sentinel
    114         # Avoid a refcycle if the target function holds an indirect

C:\Program\lib\multiprocessing\context.py in _Popen(process_obj)
    221     @staticmethod
    222     def _Popen(process_obj):
--> 223         return _default_context.get_context().Process._Popen(process_obj)
    224 
    225 class DefaultContext(BaseContext):

C:\Program\lib\multiprocessing\context.py in _Popen(process_obj)
    320         def _Popen(process_obj):
    321             from .popen_spawn_win32 import Popen
--> 322             return Popen(process_obj)
    323 
    324     class SpawnContext(BaseContext):

C:\Program\lib\multiprocessing\popen_spawn_win32.py in __init__(self, process_obj)
     63             try:
     64                 reduction.dump(prep_data, to_child)
---> 65                 reduction.dump(process_obj, to_child)
     66             finally:
     67                 set_spawning_popen(None)

C:\Program\lib\multiprocessing\reduction.py in dump(obj, file, protocol)
     58 def dump(obj, file, protocol=None):
     59     '''Replacement for pickle.dump() using ForkingPickler.'''
---> 60     ForkingPickler(file, protocol).dump(obj)
     61 
     62 #

BrokenPipeError: [Errno 32] Broken pipe

and the jupyter console output the traceback too.

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "C:\Program\lib\multiprocessing\spawn.py", line 105, in spawn_main
    exitcode = _main(fd)
  File "C:\Program\lib\multiprocessing\spawn.py", line 115, in _main
    self = reduction.pickle.load(from_parent)
AttributeError: Can't get attribute 'CAR' on <module '__main__' (built-in)>

Versions and main components

• PyMC3 Version: 3.6
• Theano Version: 1.0.3
• Python Version: 3.7
• Operating system: Win10
• How did you install PyMC3: Anaconda

[junpenglao]

This is related to #3140. For now, try

trace = pm.sample(1000, tune=1000, cores=1)

[lucianopaz]

@wangtong321, this is a typical error of the jupyter notebook with multiprocessing.

pymc3 uses multiprocessing to sample many chains in parallel, hence @junpenglao's suggestion of running sample with cores=1, in order to avoid multiprocessing entirely.

The problem is that, when multiprocessing runs with the spawn (default on windows) or forkserver protocols, it starts a completely new process to run without the class definitions run in other cells of the notebook. When the new processes spawn, they are sent through a communication pipe 1) the task that they must do and 2) the data to use. This data contains pickled instances of several classes, in particular some instance of the CAR class that you defined in a previous cell of the jupyter notebook. The thing is that the new process is unaware of CAR class definition because it was defined in a completely independent process. What ends up happending is:

1. The spawned process sees it has to unpickle a CAR instance
2. Does not find it in any loaded modules
3. Tries to look it up in the __main__ module and fails
4. Raises the AttributeError you see in the end

Regrettably, this is a problem that goes beyond pymc3, and the only workaround that you can try to do is detailed here. What you have to do is write the CAR class definition in a script outside the jupyter notebook (e.g. car_distribution.py), and in the notebook you import the CAR class as from car_distribution import CAR.

This should allow you to run the notebook with multiple cores.

[lucianopaz]

I'm hesitant on what we can do about this problem at the pymc3 level. Maybe we could add some lines explaining the potential issue of running the notebook on windows, and how to work around it. What do you think @junpenglao, @aseyboldt?

[ghost]

Thanks very much. @lucianopaz @junpenglao
My PC now has another bug : ( , I'll try your suggestion when I fix the bug.

[ghost]

OK, it worked!

[rlloretb]

Any news on this?

I am arbitrarily getting the error pipe when sampling within a Spyder cell, on MacOS.