fixed_ get_variable_name

[junpenglao]

bug fixed for issue #2224

[fonnesbeck]

Yes, that should do it. Should probably insert a repr test somewhere.

[junpenglao]

@fonnesbeck I agree, I can't think of a good way to do it though.
Also, it would be great to also add this feature to gp.

[fonnesbeck]

Yes, we should add it to GP and all the covariances, but that can wait until post-3.1.

We can test them by calling _repr_latex_ explicitly, I suppose.

[junpenglao]

But that would mean we need to test them case by case? It would increase the code size quite a lot.

[fonnesbeck]

I wouldnt test all of them, maybe just one per module to ensure that its imported and it works. Could even piggyback it on an existing test(s).

[fonnesbeck]

@junpenglao do you want me to merge this and add a test separately?

[junpenglao]

I am working on a test, something like:

with pm.Model() as model:
    x0 = pm.Binomial('Discrete', p=.5, n=10)
    x1 = pm.Normal('Continuous', mu=0., sd=1.)
    x2 = pm.GaussianRandomWalk('timeseries', mu=x1, sd=5., shape=10)
assert x0._repr_latex_(), '$Discrete \\sim \\text{Binomial}(\\mathit{n}=10, \\mathit{p}=0.5)$'
assert x1._repr_latex_(), '$Continuous \\sim \\text{Normal}(\\mathit{mu}=0.0, \\mathit{sd}=1.0)$'

However while doing so i seems to find another bug:

x2._repr_latex_()

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-17-3a78dd7af59c> in <module>()
      3     x1 = pm.Normal('Continuous', mu=0., sd=1.)
      4     x2 = pm.GaussianRandomWalk('timeseries', mu=x1, sd=5., shape=10)
----> 5 x2._repr_latex_()

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pymc3/model.py in _repr_latex_(self, name, dist)
    830         if dist is None:
    831             dist = self.distribution
--> 832         return self.distribution._repr_latex_(name=name, dist=dist)
    833 
    834     @property

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pymc3/distributions/timeseries.py in _repr_latex_(self, name, dist)
    100         sd = dist.sd
    101         return r'${} \sim \text{{GaussianRandomWalk}}(\mathit{{mu}}={}, \mathit{{sd}}={})$'.format(name,
--> 102                                                 get_variable_name(mu),
    103                                                 get_variable_name(sd))
    104 

/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/site-packages/pymc3/util.py in get_variable_name(variable)
     81     returns the argument name.
     82     """
---> 83     name = variable.name
     84     if name is None:
     85         if hasattr(variable, 'get_parents'):

AttributeError: 'float' object has no attribute 'name'

[fonnesbeck]

Good to know. We typically cast all constants passed to distributions into tensors, so apparently that's not happening in the GRW.

[junpenglao]

@fonnesbeck OK, fixing that as well.

[fonnesbeck]

Yes, it should be, for example:

self.mu = mu = tt.as_tensor_variable(mu)

[fonnesbeck]

Good to merge once tests pass.

[fonnesbeck]

Looks like a casting issue

[junpenglao]

@fonnesbeck it's the int in LKJCorr, fixing now.

[twiecki]

Is this good to merge?

[junpenglao]

@twiecki not yet, i am working on a test for _repr_latex_