baffling error-handling problem

[Chris Fonnesbeck]

I thought I knew how to do error handling in python, but apparently I
dont. I have a bunch of code to calculate statistical likelihoods, and
use error handling to catch invalid parameters. For example, for the
bernoulli distribution, I have:

def bernoulli_like( self, x, p, name='bernoulli '):
"""Bernoull i log-likelihood"""

# Ensure proper dimensionality of parameters
dim = shape(x)
p = resize(p,dim)

# Ensure valid values of parameters
if sum(p>=1 or p<=0): raise LikelihoodError

... etc.

where LikelihoodError is simply a subclass of ValueError that I created:

class LikelihoodError (ValueError):
"Log-likelihood is invalid or negative infinite"
I catch these errors with the following:

try:
like = self.calculate_ likelihood()
except LikelihoodError :
return 0

So far, so good. When I calculate_likel ihood is called in the above,
which contains a call to bernoulli_like:

p = invlogit(beta0 + sum([b*h for b,h in zip(self.beta,h ab)]))

like=self.berno ulli_like(x,p)

I get the following when an invalid parameter is passed:

Traceback (most recent call last):
File "C:\Conroy\work ing\resource_se lection_ms\anal yses\IIbq\sampl ed\new_chris\mo del_000.py",
line 381, in ?
model.sample(it erations=iter, burn=burn,plot= False)
File "C:\Python23\Li b\site-packages\PyMC\M CMC.py", line 1691, in sample
self._like = self.calculate_ likelihood()
File "C:\Conroy\work ing\resource_se lection_ms\anal yses\IIbq\sampl ed\new_chris\mo del_000.py",
line 194, in calculate_likel ihood
like+=self.bern oulli_like(x,p)
File "C:\Python23\Li b\site-packages\PyMC\M CMC.py", line 868, in bernoulli_like
if sum(p>=1 or p<=0): raise LikelihoodError
LikelihoodError
I have no idea how this can happen, given how I have coded this.
Anyone see what I must be missing?

Thanks,
C.

[Michael Hudson]

Chris Fonnesbeck <fo********@gma il.com> writes:

I thought I knew how to do error handling in python, but apparently I
dont. I have a bunch of code to calculate statistical likelihoods, and
use error handling to catch invalid parameters. For example, for the
[...]
bernoulli distribution, I have:
I have no idea how this can happen, given how I have coded this.
Anyone see what I must be missing?

Is it possible you have two classes called LikelihoodError ? One in
__main__, one in some_module_of_ yours, maybe.

Cheers,
mwh

--
<bruce> how are the jails in israel?
<itamar> well, the one I was in was pretty nice
-- from Twisted.Quotes

[Scott David Daniels]

Chris Fonnesbeck wrote:

I thought I knew how to do error handling in python, but apparently I
dont. I have a bunch of code to calculate statistical likelihoods, and
use error handling to catch invalid parameters. For example, for the
bernoulli distribution, I have:

def bernoulli_like( self, x, p, name='bernoulli '):
... if sum(p>=1 or p<=0): raise LikelihoodError ...

where LikelihoodError is simply a subclass of ValueError that I created:

class LikelihoodError (ValueError):
"Log-likelihood is invalid or negative infinite"

I catch these errors with the following:

try: like = self.calculate_ likelihood()
except LikelihoodError :
return 0

... [when using] ...
like=self.berno ulli_like(x,p)

I get the following when an invalid parameter is passed:

Traceback (most recent call last):
File "...\model_000. py", line 381, in ?
model.sample(it erations=iter, burn=burn,plot= False)
File "...\PyMC\MCMC. py", line 1691, in sample
self._like = self.calculate_ likelihood()
File "...\model_000. py", line 194, in calculate_likel ihood
like+=self.bern oulli_like(x,p)
File "...\MCMC.p y", line 868, in bernoulli_like
if sum(p>=1 or p<=0): raise LikelihoodError
LikelihoodError

I have no idea how this can happen, given how I have coded this.

Might you be referring to a different LikelihoodError in the
try: ... except ... part of your code than in the ... raise ... part?
Similarly defined classes are not the same class. If you didn't
get LikelihoodError in mode4l_000.py with the moral equivalent of
from MCMC import LikelihoodError
then this is what is going wrong.
By the way, if it were I, I'd: raise LikelihoodError (p) just so I
could discover a bit of what went wrong.

--Scott David Daniels
Sc***********@A cm.Org