Compilation failed (return status=1)

[mick-phemex]

Description of your problem

To begin with, I have followed up the instruction of the installation guide and deploy the new virtual environment with the yml file presented over that page, but I still met with the same problem as I executed the following codes. I have tried out many ways to solve the problem, please assist me to solve the problem.

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

import pymc3 as pm
import numpy as np
import pandas as pd
import scipy.stats as stats
from datetime import datetime
import theano.tensor as T

early = 10
late = 12
y = np.r_[np.random.poisson(early, 25), np.random.poisson(late, 75)]
      
niter = 10000
t = range(len(y))
with pm.Model() as change_point:
    cp = pm.DiscreteUniform('change_point', lower=0, upper=len(y), testval=len(y)//2)
    mu0 = pm.Exponential('mu0', 1/y.mean())
    mu1 = pm.Exponential('mu1', 1/y.mean())
    mu = T.switch(t < cp, mu0, mu1)
    Y_obs = pm.Poisson('Y_obs', mu=mu, observed=y)
    trace = pm.sample(niter)
pm.traceplot(trace, varnames=['change_point', 'mu0', 'mu1'])

Please provide the full traceback.

Complete error traceback

[You can find the C code in this temporary file: C:\Users\Mick\AppData\Local\Temp\theano_compilation_error_xk8zcr1g
Traceback (most recent call last):

  File "C:\....py", line 48, in <module>
    mu = T.switch(t < cp, mu0, mu1)

  File "C:\...\lib\site-packages\theano\tensor\var.py", line 41, in __gt__
    rval = theano.tensor.basic.gt(self, other)

  File "C:\...\lib\site-packages\theano\graph\op.py", line 253, in __call__
    compute_test_value(node)

  File "C:\...\lib\site-packages\theano\graph\op.py", line 126, in compute_test_value
    thunk = node.op.make_thunk(node, storage_map, compute_map, no_recycling=[])

  File "C:\...\lib\site-packages\theano\graph\op.py", line 634, in make_thunk
    return self.make_c_thunk(node, storage_map, compute_map, no_recycling)

  File "C:\...\lib\site-packages\theano\graph\op.py", line 600, in make_c_thunk
    outputs = cl.make_thunk(

  File "C:\...\lib\site-packages\theano\link\c\basic.py", line 1203, in make_thunk
    cthunk, module, in_storage, out_storage, error_storage = self.__compile__(

  File "C:\...\lib\site-packages\theano\link\c\basic.py", line 1138, in __compile__
    thunk, module = self.cthunk_factory(

  File "C:\...\lib\site-packages\theano\link\c\basic.py", line 1634, in cthunk_factory
    module = get_module_cache().module_from_key(key=key, lnk=self)

  File "C:\...\lib\site-packages\theano\link\c\cmodule.py", line 1191, in module_from_key
    module = lnk.compile_cmodule(location)

  File "C:\...\lib\site-packages\theano\link\c\basic.py", line 1543, in compile_cmodule
    module = c_compiler.compile_str(

  File "C:\...\lib\site-packages\theano\link\c\cmodule.py", line 2546, in compile_str
    raise Exception(

Exception: ('Compilation failed (return status=1): C:\\...\\AppData\\Local\\Temp\\ccujaONv.s: Assembler messages:\r. C:\\...\\AppData\\Local\\Temp\\ccujaONv.s:89: Error: invalid register for .seh_savexmm\r. ', 'FunctionGraph(Elemwise{gt,no_inplace}(<TensorType(int64, (True,))>, TensorConstant{[ 0  1  2 .. 97 98 99]}))')]

Please provide any additional information below.

Versions and main components

• PyMC3 Version: v3.11.2
• Aesara/Theano Version: v1.1.2
• Python Version: 3.8
• Operating system: win10
• How did you install PyMC3: conda (with yml file consistent to the setting from here

[michaelosthege]

Can you run a simpler model that doesn't involve theano.tensor.switch?

Some more info that might help:

• Windows 10 64 bit I assume?
• Are you working with Anaconda or miniconda?
• What does your PATH environment variable contain?

[mick-phemex]

The following example is another model that I tried, the same error still happened. I assumed that the error code is this one:

start = pm.find_MAP()

np.random.seed(123)
n = 11
_a = 6
_b = 2
x = np.linspace(0, 1, n)
y = _a*x + _b + np.random.randn(n)
niter = 10000
with pm.Model() as linreg:
    a = pm.Normal('a', mu=0, sd=100)
    b = pm.Normal('b', mu=0, sd=100)
    sigma = pm.HalfNormal('sigma', sd=1)

    y_est = a*x + b
    likelihood = pm.Normal('y', mu=y_est, sd=sigma, observed=y)

    start = pm.find_MAP()
    step = pm.NUTS(scaling=start)
    trace = pm.sample(niter, step, start, random_seed=123)
t = trace[niter//2:]
pm.traceplot(trace, varnames=['a', 'b'])

Additional info:

• Yes, it is Windows 10 64 bit.
• I am working with Anaconda.
• May I ask specifically what you mean about PATH environment variable? user variables or system variables? I assume you meant system PATH environment variables, it contains the following pathes.

Thank you for your response! @michaelosthege

[michaelosthege]

I can't tell for sure (and I won't risk my installations trying to validate), but maybe the C:\MinGW\bin in your PATH is (part of) the problem.
Maybe it takes precedence over the MinGW from the environment?

For comparison this is my user PATH environment variable:

In an activated PyMC3 environment I end up with this:

>>> print(os.getenv("PATH").replace(";", "\n"))
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3\Library\mingw-w64\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3\Library\usr\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3\Library\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3\Scripts
C:\Users\osthege\AppData\Local\Continuum\miniconda3\envs\pm3v3\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\condabin
C:\Program Files\AdoptOpenJDK\jre-8.0.292.10-hotspot\bin
C:\Program Files\Microsoft MPI\Bin
C:\WINDOWS\system32
C:\WINDOWS
C:\WINDOWS\System32\Wbem
C:\WINDOWS\System32\WindowsPowerShell\v1.0
C:\Program Files\dotnet
C:\Program Files\Microsoft SQL Server\130\Tools\Binn
C:\WINDOWS\System32\OpenSSH
C:\Program Files (x86)\MySQL\MySQL Utilities 1.6
C:\Program Files\Microsoft SQL Server\Client SDK\ODBC\170\Tools\Binn
C:\Program Files\Git\cmd
C:\Program Files\PuTTY
C:\Program Files (x86)\NVIDIA Corporation\PhysX\Common
C:\Program Files\NVIDIA Corporation\NVIDIA NvDLISR
C:\Program Files\Docker\Docker\resources\bin
C:\ProgramData\DockerDesktop\version-bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3
C:\Users\osthege\AppData\Local\Continuum\miniconda3\Library\mingw-w64\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\Library\usr\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\Library\bin
C:\Users\osthege\AppData\Local\Continuum\miniconda3\Scripts
C:\Users\osthege\AppData\Local\Microsoft\WindowsApps
C:\Users\osthege\AppData\Local\GitHubDesktop\bin
C:\Users\osthege\.dotnet\tools
C:\Users\osthege\AppData\Local\Programs\Microsoft VS Code\bin
C:\Users\osthege\AppData\Local\Microsoft\WindowsApps

[sreedat]

@michaelosthege thanks for posting this configuration of your User variables. As soon as I set up mine correctly, all my problems have been resolved. To test if the RTools set up was conflicting with my Anaconda / Pymc3 compilation, I removed RTools completely and when I ran import theano as tp, it gave me a cannot find g++ error. With m2w64 tool chain installed this error should not have occurred.
So I set up my User variables per your post, and disabled by RTools installation by changing my RTools folder name. Now all of my Pymc3 and Bambi models are compiling correctly.

[michaelosthege]

@sreedat thanks for that confirmation! I guess we should add corresponding troubleshooting instructions to the install guide..

[sreedat]

@michaelosthege As a token of my appreciation I would like to create this troubleshooting guide and help developers such as you focus on the developmental aspects of these packages. During the past week, I went through installation of Pymc3 and testing it on Ubuntu 20.04 LTS, Windows 8.1 and Windows 10. So I can provide these instructions from a comparison perspective too if needed. I have never done any thing like this before. So if you can point me to a guide on how to do this, I would love to help. You helped slow down my balding without a doubt, so a double thank you!

[michaelosthege]

@michaelosthege As a token of my appreciation I would like to create this troubleshooting guide and help developers such as you focus on the developmental aspects of these packages. During the past week, I went through installation of Pymc3 and testing it on Ubuntu 20.04 LTS, Windows 8.1 and Windows 10. So I can provide these instructions from a comparison perspective too if needed. I have never done any thing like this before. So if you can point me to a guide on how to do this, I would love to help. You helped slow down my balding with a doubt, so a double thank you!

Having been through this in the past I can understand your pain 😅
As always contributions are welcome!
We have the guide in a Wiki page because this way project members can quickly update it without going through the pull requests.
The downside is that it's not as straightforward to take outside contributions.

I see two ways how you could contribute to the wiki page:

• Create a New Issue, copy the wiki page content and edit the markdown over in the new issue. Then we'll copy it to the wiki manually.
• According to this stackoverflow post it is possible to edit the Wiki with a git workflow, but it doesn't look very straightforward. Wouldn't recommend unless you feel confident with git.

To focus the reader attention we could also just link from a troubleshooting section to Github issues where we can have more information and discussion.