import error on test code

[YLouWashU]

I get this import error when importing the eptools_ext.so into ipython:

ImportError Traceback (most recent call last)
/usr/lib/python2.7/dist-packages/IPython/utils/py3compat.pyc in execfile(fname, *where)
176 else:
177 filename = fname
--> 178 builtin.execfile(filename, *where)

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/test/binclass/eptest_binclass.py in ()
18 import time # Profiling
19
---> 20 import apbsint as abt
21
22 # Helper functions

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/init.py in ()
12
13 from apbsint.coup_fact import *
---> 14 from apbsint.utilities import *
15 from apbsint.inference import *

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/utilities.py in ()
18 import apbsint.helpers as helpers
19 import apbsint.coup_fact as cf
---> 20 import apbsint.eptools_ext as epx
21 import apbsint.ptannotate_ext as pta
22

ImportError: /home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/eptools_ext.so: undefined symbol: eptwrap_fact_compmarginals_bvprec

I checked compilation log, there seems to be no abnormality there. So I guess this could be a problem with linking?

[mseeger]

Hello,

first thanks for your interest in the project. You are the first external
user.

Since I have not anticipated other users, I have been careless with the
branching. The master branch right now
is not stable, i.e. does not build right now.

Knowing now that somebody else is playing with the code, I will try and be
more careful. I will try to get master
into a stable state and do new development in another branch. I can notify
you once this is done.

Another complication with this project is that for some new developments, I
am using GPL licensed code (the GNU
scientific library). I cannot include the code that makes use of GSL, due
to the viral license. I have not found a good
solution to this problem: right now, I keep the GPL-dependent part private.
However, the test case eptest_binclass.py
runs without this extension (once I cleaned up the master branch).

I hope you are still interested and may at some point be interested in
contributing.

Bye,

On Fri, Mar 21, 2014 at 9:57 PM, YLouWashU notifications@github.com wrote:

I get this import error when importing the eptools_ext.so into ipython:

ImportError Traceback (most recent call last)
/usr/lib/python2.7/dist-packages/IPython/utils/py3compat.pyc in
execfile(fname, where)
176 else:
177 filename = fname
--> 178 *builtin.execfile(filename, *where)

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/test/binclass/eptest_binclass.py
in ()
18 import time # Profiling
19
---> 20 import apbsint as abt
21
22 # Helper functions

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/
init.py in ()
12
13 from apbsint.coup_fact import *
---> 14 from apbsint.utilities import *
15 from apbsint.inference import *

/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/utilities.py
in ()
18 import apbsint.helpers as helpers
19 import apbsint.coup_fact as cf
---> 20 import apbsint.eptools_ext as epx
21 import apbsint.ptannotate_ext as pta
22

ImportError:
/home/louyang/PAT_HomeDir/MachineLearningProj/apbsint-master/python/apbsint/eptools_ext.so:
undefined symbol: eptwrap_fact_compmarginals_bvprec

I checked compilation log, there seems to be no abnormality there. So I
guess this could be a problem with linking?

Reply to this email directly or view it on GitHubhttps://github.com//issues/1
.

Matthias Seeger

[YLouWashU]

Thank you so much for the work! I'll keep on playing around with the code and see if I can contribute. And I have a quick question regarding the Bayesian Inference Approximation algorithm (in your 2010 MR in Medicine paper). So if I am understanding correctly, the posterior is approximated by a Gaussian fit, which is parameterized by a vector Gamma denoting the variances of Gaussian kernels (1D Gaussian each?). So the output should be this Gamma vector, so these 1D Gaussian kernels are independent of each other? So the approximated posterior is a product of these Gaussian kernels? Thanks a lot!

[mseeger]

Hello,

yes the goal is to find the optimal gamma parameters. While this is a
product, the resulting distribution is not
factorizing, because each Gaussian "kernel" itself couples all components
of u.

Note that the new toolbox does not contain that algorithm. For that, you
need to use glm-ie of my ex-student
Hannes Nickisch. The new toolbox is for EP, as I've been using in earlier
work, or as is implemented in infer.NET.

Bye,

On Sat, Mar 22, 2014 at 5:17 PM, YLouWashU notifications@github.com wrote:

Thank you so much for the work! I'll keep on playing around with the code
and see if I can contribute. And I have a quick question regarding the
Bayesian Inference Approximation algorithm (in your 2010 MR in Medicine
paper). So if I am understanding correctly, the posterior is approximated
by a Gaussian fit, which is parameterized by a vector Gamma denoting the
variances of Gaussian kernels (1D Gaussian each?). So the output should be
this Gamma vector, so these 1D Gaussian kernels are independent of each
other? So the approximated posterior is a product of these Gaussian
kernels? Thanks a lot!

Reply to this email directly or view it on GitHubhttps://github.com//issues/1#issuecomment-38355780
.

Matthias Seeger

[mseeger]

Hello,

the master branch is stable now (please only use that one). I've created
the dev branch for my recent changes,
this is unstable and should not be used.

Bye,

On Sat, Mar 22, 2014 at 5:17 PM, YLouWashU notifications@github.com wrote:

Thank you so much for the work! I'll keep on playing around with the code
and see if I can contribute. And I have a quick question regarding the
Bayesian Inference Approximation algorithm (in your 2010 MR in Medicine
paper). So if I am understanding correctly, the posterior is approximated
by a Gaussian fit, which is parameterized by a vector Gamma denoting the
variances of Gaussian kernels (1D Gaussian each?). So the output should be
this Gamma vector, so these 1D Gaussian kernels are independent of each
other? So the approximated posterior is a product of these Gaussian
kernels? Thanks a lot!

Reply to this email directly or view it on GitHubhttps://github.com//issues/1#issuecomment-38355780
.

Matthias Seeger

[YLouWashU]

Thank you so much! I'll look into both apbsint and glm-ie to see which one better suits my need!