Today

• 10 hours ago

  Testimonials Wiki page commented on by alexandre.birolleau   -   It would be nice to say in the installation part that it is necessary to have python-dev to install pymc. Thanks. Alexandre Birolleau

  It would be nice to say in the installation part that it is necessary to have python-dev to install pymc. Thanks. Alexandre Birolleau

Last 7 days

• Dec 28, 2009

  issue 298 (HDF5 backend ignores dbcomplib and dbcomplevel) reported by todd.andrew.small   -   What steps will reproduce the problem? 1. Create a MCMC instance, e.g., M = pymc.MCMC(<some model>, db='hdf5', dbcomplib='zlib', dbcomplevel=1). 2. Sample from the model. 3. M.db.close() 4. Confirm with h5stat that no GZIP filters have been used. What version of the product are you using? On what operating system? PyMC SVN version installed on 12/18/2009, Python 2.4.6, OS X 10.6.2, PyTables 2.1.2 Please provide any additional information below. The bug seems to be in line ~227 of hdf5.py. At least with my version of PyTables, self._h5file has an attribute "filters", and so the line self.filter = getattr(self._h5file, 'filters', tables.Filters(complevel=dbcomplevel, complib=dbcomplib) doesn't actually assign the compression filter to self._h5file. One simple fix is to create the filter and then open the file as follows: self.filter = tables.Filters(complevel=dbcomplevel, complib=dbcomplib) self._h5file = tables.openFile(self.dbname, self.mode, filters=self.filter)

  What steps will reproduce the problem? 1. Create a MCMC instance, e.g., M = pymc.MCMC(<some model>, db='hdf5', dbcomplib='zlib', dbcomplevel=1). 2. Sample from the model. 3. M.db.close() 4. Confirm with h5stat that no GZIP filters have been used. What version of the product are you using? On what operating system? PyMC SVN version installed on 12/18/2009, Python 2.4.6, OS X 10.6.2, PyTables 2.1.2 Please provide any additional information below. The bug seems to be in line ~227 of hdf5.py. At least with my version of PyTables, self._h5file has an attribute "filters", and so the line self.filter = getattr(self._h5file, 'filters', tables.Filters(complevel=dbcomplevel, complib=dbcomplib) doesn't actually assign the compression filter to self._h5file. One simple fix is to create the filter and then open the file as follows: self.filter = tables.Filters(complevel=dbcomplevel, complib=dbcomplib) self._h5file = tables.openFile(self.dbname, self.mode, filters=self.filter)
• Dec 24, 2009

  issue 247 (Implement special methods (__float__, __len__, __int__) for ...) commented on by anand.prabhakar.patil   -   shoot, i don't know how I missed that. Now I'm stuck at DFW waiting for a flight without my laptop so I can't really fix it. I'll get it in the new year though.

  shoot, i don't know how I missed that. Now I'm stuck at DFW waiting for a flight without my laptop so I can't really fix it. I'll get it in the new year though.

Last 30 days

• Dec 23, 2009

  issue 247 (Implement special methods (__float__, __len__, __int__) for ...) commented on by david.huard   -   A lot of tests in test_special_methods raise an error. Also, it is not exercised in pymc.test(). Is this normal/voluntary/expected ?

  A lot of tests in test_special_methods raise an error. Also, it is not exercised in pymc.test(). Is this normal/voluntary/expected ?
• Dec 23, 2009

  Compilation (How to compile PyMC at home) Wiki page edited by david.huard   -   Revision r1391 Edited wiki page through web user interface.

  Revision r1391 Edited wiki page through web user interface.
• Dec 23, 2009

  Compilation (How to compile PyMC at home) Wiki page edited by david.huard   -   Revision r1390 Edited wiki page through web user interface.

  Revision r1390 Edited wiki page through web user interface.
• Dec 23, 2009

  Compilation (How to compile PyMC at home) Wiki page added by david.huard   -   Revision r1389 Created wiki page through web user interface.

  Revision r1389 Created wiki page through web user interface.
• Dec 23, 2009

  issue 297 (install error on ubuntu9.10) Status changed by david.huard   -  

  Status: Invalid

  Status: Invalid
• Dec 22, 2009

  issue 297 (install error on ubuntu9.10) commented on by disgrid   -   Dear David, That's it! Thank you so much!!! Frank.Xu

  Dear David, That's it! Thank you so much!!! Frank.Xu
• Dec 22, 2009

  issue 246 (Docs changes for 2.1) Status changed by david.huard   -   I added a note before 3.4.

  Status: Fixed

  I added a note before 3.4.

  Status: Fixed
• Dec 22, 2009

  issue 289 (Special methods causing stack overflows) Status changed by david.huard   -  

  Status: Verified

  Status: Verified
• Dec 22, 2009

  issue 276 (Initialization of variables fails for some parameter shapes) Status changed by david.huard   -   Looks good.

  Status: Verified

  Looks good.

  Status: Verified
• Dec 22, 2009

  issue 297 (install error on ubuntu9.10) commented on by david.huard   -   Have you installed python-dev ?

  Have you installed python-dev ?
• Dec 22, 2009

  issue 297 (install error on ubuntu9.10) reported by disgrid   -   What steps will reproduce the problem? 1.python setup.py build 2. 3. What is the expected output? What do you see instead? error: Command "gcc -pthread -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -fPIC -Ibuild/src.linux-i686-2.6 -I/usr/lib/python2.6/dist-packages/numpy/core/include -I/usr/include/python2.6 -c build/src.linux-i686-2.6/fortranobject.c -o build/temp.linux-i686-2.6/build/src.linux-i686-2.6/fortranobject.o" failed with exit status 1 What version of the product are you using? On what operating system? pymc2.0 ubuntu9.10 Please provide any additional information below. I was trying to install pymc but i met the error. I have installed gfortran gcc and numpy by apt-get install gfortran gcc numpy. Can anyone help me?

  What steps will reproduce the problem? 1.python setup.py build 2. 3. What is the expected output? What do you see instead? error: Command "gcc -pthread -fno-strict-aliasing -DNDEBUG -g -fwrapv -O2 -Wall -Wstrict-prototypes -fPIC -Ibuild/src.linux-i686-2.6 -I/usr/lib/python2.6/dist-packages/numpy/core/include -I/usr/include/python2.6 -c build/src.linux-i686-2.6/fortranobject.c -o build/temp.linux-i686-2.6/build/src.linux-i686-2.6/fortranobject.o" failed with exit status 1 What version of the product are you using? On what operating system? pymc2.0 ubuntu9.10 Please provide any additional information below. I was trying to install pymc but i met the error. I have installed gfortran gcc and numpy by apt-get install gfortran gcc numpy. Can anyone help me?
• Dec 19, 2009

  issue 294 (Refactor GP package's PyMC objects) commented on by anand.prabhakar.patil   -   One more thing: need fast_path multi-eval function: fast_eval(M, C, x, mean=False, full_covariance=False, cholesky=False, diag=False, nugget=None)

  One more thing: need fast_path multi-eval function: fast_eval(M, C, x, mean=False, full_covariance=False, cholesky=False, diag=False, nugget=None)
• Dec 18, 2009

  pymc-2.1alpha.win32-py2.5.exe (PyMC 2.1 (alpha) installer for Windows XP) file uploaded by fonnesbeck   -  

  Labels: Type-Installer OpSys-Windows Featured

  Labels: Type-Installer OpSys-Windows Featured
• Dec 16, 2009

  issue 296 (Specifying path for graph.graph fails) Status changed by fonnesbeck   -   Das ist gut.

  Status: Fixed

  Das ist gut.

  Status: Fixed
• Dec 16, 2009

  issue 296 (Specifying path for graph.graph fails) commented on by anand.prabhakar.patil   -   Now?

  Now?
• Dec 16, 2009

  issue 296 (Specifying path for graph.graph fails) commented on by fonnesbeck   -   The test gives the following failures; note how I specified the path argument: ====================================================================== ERROR: test_graph (pymc.tests.test_graph.test_graph) ---------------------------------------------------------------------- Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 41, in test_graph pm.graph.graph(M, path="testresults", **dict(args)) File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 266, in graph name = name + '.' + ext UnboundLocalError: local variable 'ext' referenced before assignment ====================================================================== ERROR: test_moral (pymc.tests.test_graph.test_graph) ---------------------------------------------------------------------- Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 51, in test_moral pm.graph.moral_graph(M, path="testresults") File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 58, in moral_graph name = name + '.' + ext UnboundLocalError: local variable 'ext' referenced before assignment

  The test gives the following failures; note how I specified the path argument: ====================================================================== ERROR: test_graph (pymc.tests.test_graph.test_graph) ---------------------------------------------------------------------- Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 41, in test_graph pm.graph.graph(M, path="testresults", **dict(args)) File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 266, in graph name = name + '.' + ext UnboundLocalError: local variable 'ext' referenced before assignment ====================================================================== ERROR: test_moral (pymc.tests.test_graph.test_graph) ---------------------------------------------------------------------- Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 51, in test_moral pm.graph.moral_graph(M, path="testresults") File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 58, in moral_graph name = name + '.' + ext UnboundLocalError: local variable 'ext' referenced before assignment
• Dec 16, 2009

  issue 296 (Specifying path for graph.graph fails) Status changed by anand.prabhakar.patil   -   Now?

  Status: NeedsReview

  Now?

  Status: NeedsReview
• Dec 16, 2009

  issue 296 (Specifying path for graph.graph fails) reported by fonnesbeck   -   I was trying to specify a path for plots in test_graph.py to keep all the test output in ./testresults, but this fails: Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 51, in test_moral pm.graph.moral_graph(M, path="testresults") File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 57, in moral_graph model.moral_dot_object.write(path=path, format=format, prog=prog) File "build/bdist.macosx-10.6-universal/egg/pydot.py", line 1692, in write dot_fd = file(path, "w+b") IOError: [Errno 21] Is a directory: 'testresults' Seems like graph.graph's use of path differs from other plotting functions in PyMC.

  I was trying to specify a path for plots in test_graph.py to keep all the test output in ./testresults, but this fails: Traceback (most recent call last): File "/Users/fonnesbeck/Code/pymc/pymc/tests/test_graph.py", line 51, in test_moral pm.graph.moral_graph(M, path="testresults") File "/Users/fonnesbeck/Code/pymc/pymc/graph.py", line 57, in moral_graph model.moral_dot_object.write(path=path, format=format, prog=prog) File "build/bdist.macosx-10.6-universal/egg/pydot.py", line 1692, in write dot_fd = file(path, "w+b") IOError: [Errno 21] Is a directory: 'testresults' Seems like graph.graph's use of path differs from other plotting functions in PyMC.
• Dec 14, 2009

  issue 286 (Making loops more usable) commented on by anand.prabhakar.patil   -   You can do @deterministic(plot=False, trace=False, name='m_%i'%i) def m_i(m1=m1, m2=m2, w=w_i): """Appropriate group mean""" if w: return m2 return m1 and skip the m_i.__name__ bit. But it could be made easier, sure. If you automate this, and tag the variables as you create them, you could make issue 270 easier to deal with.

  You can do @deterministic(plot=False, trace=False, name='m_%i'%i) def m_i(m1=m1, m2=m2, w=w_i): """Appropriate group mean""" if w: return m2 return m1 and skip the m_i.__name__ bit. But it could be made easier, sure. If you automate this, and tag the variables as you create them, you could make issue 270 easier to deal with.
• Dec 14, 2009

  issue 270 (graph.dag() breaks down with containers) commented on by anand.prabhakar.patil   -   To make this work reliably I think you'd have to tag variables as members of plates at creation. Currently, if you create a bunch of variables using a loop, there's no record of that afterward.

  To make this work reliably I think you'd have to tag variables as members of plates at creation. Currently, if you create a bunch of variables using a loop, there's no record of that afterward.
• Dec 14, 2009

  issue 294 (Refactor GP package's PyMC objects) Labels changed by anand.prabhakar.patil   -  

  Labels: Milestone-Release2.1 −Milestone-Release2.2

  Labels: Milestone-Release2.1 −Milestone-Release2.2
• Dec 14, 2009

  issue 294 (Refactor GP package's PyMC objects) Status changed by anand.prabhakar.patil   -   Done in my personal branch, will merge into the trunk in the first release candidate.

  Status: NeedsReview

  Done in my personal branch, will merge into the trunk in the first release candidate.

  Status: NeedsReview
• Dec 10, 2009

  Benchmarks Wiki page edited by anand.prabhakar.patil   -   Revision r1388 Edited wiki page through web user interface.

  Revision r1388 Edited wiki page through web user interface.
• Dec 08, 2009

  issue 295 (Unladen Swallow) reported by anand.prabhakar.patil   -   After 2.1, I'm going to try to make a branch targeting Unladen Swallow. I'll take out the low-level Pyrex bits and try to optimize the basic objects for their new context.

  After 2.1, I'm going to try to make a branch targeting Unladen Swallow. I'll take out the low-level Pyrex bits and try to optimize the basic objects for their new context.
• Dec 08, 2009

  issue 32 (Gibbs samplers) Owner changed by anand.prabhakar.patil   -  

  Owner: ---

  Owner: ---
• Dec 08, 2009

  issue 260 (git-svn robot crashing) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 235 (rtruncnorm could be improved) Owner changed by anand.prabhakar.patil   -  

  Owner: ---

  Owner: ---
• Dec 08, 2009

  issue 203 (Log-error function for truncnorm_like) Owner changed by anand.prabhakar.patil   -  

  Owner: ---

  Owner: ---
• Dec 08, 2009

  issue 159 (Fortran likelihoods need to check shapes of input arguments) Owner changed by anand.prabhakar.patil   -  

  Owner: ---

  Owner: ---
• Dec 08, 2009

  issue 294 (Refactor GP package's PyMC objects) reported by anand.prabhakar.patil   -   Currently it's overcomplicated and slow. Should have the following object model: C -> C_eval -> S_eval M -> M_eval M_eval, S_eval -> f_eval so now f_eval is a simple multivariate normal, and M, C, S_eval -> M_obs, C_obs (will need to make observe accept a pre-computed Cholesky factor) M_obs, C_obs -> f Now f should be handled by the existing objects, but it will always have no mesh, so they can be drastically simplified.

  Currently it's overcomplicated and slow. Should have the following object model: C -> C_eval -> S_eval M -> M_eval M_eval, S_eval -> f_eval so now f_eval is a simple multivariate normal, and M, C, S_eval -> M_obs, C_obs (will need to make observe accept a pre-computed Cholesky factor) M_obs, C_obs -> f Now f should be handled by the existing objects, but it will always have no mesh, so they can be drastically simplified.
• Dec 08, 2009

  issue 257 (Google Code now supports Mercurial) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 252 (Launchpad's PPA auto-builds debs) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 245 (Customized syntax) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 210 (Should MultiModelInference be included in the release?) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 167 (Parser for WinBugs models) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 162 (Parsing R model formula syntax) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 147 ('Time' flag in variables for profiling) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 127 (making 'p' an infinite generator in Categorical and Multinom...) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 108 (Ravelling groups of stochastics) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 08, 2009

  issue 56 (Derivatives) Status changed by anand.prabhakar.patil   -  

  Status: WontFix

  Status: WontFix
• Dec 07, 2009

  issue 293 (MCMC.dic() broken) Status changed by fonnesbeck   -   OK, fixed this, and made DIC a property just like deviance: In [1]: from pymc import * In [2]: from pymc.examples import model_1 In [3]: M = MCMC(model_1) In [4]: M.sample(10000, 5000) In [5]: M.dic Out[5]: 369.58192113078633 http://github.com/pymc-devs/pymc/commit/392685cf7824f760bc8efb2b73c30c5e15e10a4c

  Status: NeedsReview

  OK, fixed this, and made DIC a property just like deviance: In [1]: from pymc import * In [2]: from pymc.examples import model_1 In [3]: M = MCMC(model_1) In [4]: M.sample(10000, 5000) In [5]: M.dic Out[5]: 369.58192113078633 http://github.com/pymc-devs/pymc/commit/392685cf7824f760bc8efb2b73c30c5e15e10a4c

  Status: NeedsReview
• Dec 07, 2009

  issue 293 (MCMC.dic() broken) commented on by fonnesbeck   -   I can confirm (at least on the sqlite backend) that deviance is being tallied every iteration, so it appears that it is just the reporting that is screwed up.

  I can confirm (at least on the sqlite backend) that deviance is being tallied every iteration, so it appears that it is just the reporting that is screwed up.
• Dec 07, 2009

  issue 243 (ZeroProbability error in Dirichlet) Status changed by fonnesbeck   -   Closing.

  Status: Fixed

  Closing.

  Status: Fixed
• Dec 07, 2009

  issue 243 (ZeroProbability error in Dirichlet) commented on by anand.prabhakar.patil   -   S'OK?

  S'OK?
• Dec 07, 2009

  issue 293 (MCMC.dic() broken) commented on by anand.prabhakar.patil   -   Yeah, MCMC now has a _funs_to_tally attribute that lets it tally the return value of any old function. In [2]: N = Normal('N',0,1) In [3]: M = MCMC([N]) In [4]: M._funs_to_tally Out[4]: {'N': <bound method Normal.get_value of <pymc.distributions.Normal 'N' at 0x106b68910>>, 'deviance': <bound method MCMC._sum_deviance of <pymc.MCMC.MCMC object at 0x106b68f50>>} I'd prefer it if we routed DIC through that rather than making it a deterministic.

  Yeah, MCMC now has a _funs_to_tally attribute that lets it tally the return value of any old function. In [2]: N = Normal('N',0,1) In [3]: M = MCMC([N]) In [4]: M._funs_to_tally Out[4]: {'N': <bound method Normal.get_value of <pymc.distributions.Normal 'N' at 0x106b68910>>, 'deviance': <bound method MCMC._sum_deviance of <pymc.MCMC.MCMC object at 0x106b68f50>>} I'd prefer it if we routed DIC through that rather than making it a deterministic.
• Dec 06, 2009

  issue 293 (MCMC.dic() broken) reported by fonnesbeck   -   DIC calculation in MCMC is broken because for some reason the deviance attribute is a float rather than a Node. Specifying calc_deviance=True (the default) when initializing MCMC is supposed to tally the deviance at each iteration, but it appears not to.

  DIC calculation in MCMC is broken because for some reason the deviance attribute is a float rather than a Node. Specifying calc_deviance=True (the default) when initializing MCMC is supposed to tally the deviance at each iteration, but it appears not to.
• Dec 04, 2009

  issue 55 (Implement GOF methods) Status changed by fonnesbeck   -   We already have some GOF (see section 7.3 of the paper). Its perhaps not as automated as we ultimately want, but it will certainly suffice for now. I say close this, and open a new ticket when we get the urge to improve it in the future.

  Status: Fixed

  We already have some GOF (see section 7.3 of the paper). Its perhaps not as automated as we ultimately want, but it will certainly suffice for now. I say close this, and open a new ticket when we get the urge to improve it in the future.

  Status: Fixed