for Bounded RV: 'HalfNormal' object has no attribute 'median'

[aakhmetz]

I use a simple code:

with pm.Model() as model:
    D = pm.HalfNormal('D',1)
    S = pm.Bound(pm.HalfNormal, upper=D)('S', 1)

Two half normals, one of which is bounded, then the default parameter value in S is not set to sd or tau as one may expect. The code above gives an error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-30-1331a459e0b2> in <module>
      1 with pm.Model() as model:
      2     D = pm.HalfNormal('D',1)
----> 3     S = pm.Bound(pm.HalfNormal, upper=D)('S', 1)

~/anaconda/lib/python3.7/site-packages/pymc3/distributions/bound.py in __call__(self, name, *args, **kwargs)
    217         if issubclass(self.distribution, Continuous):
    218             return _ContinuousBounded(name, self.distribution,
--> 219                                       self.lower, self.upper, *args, **kwargs)
    220         elif issubclass(self.distribution, Discrete):
    221             return _DiscreteBounded(name, self.distribution,

~/anaconda/lib/python3.7/site-packages/pymc3/distributions/distribution.py in __new__(cls, name, *args, **kwargs)
     39                 raise TypeError("observed needs to be data but got: {}".format(type(data)))
     40             total_size = kwargs.pop('total_size', None)
---> 41             dist = cls.dist(*args, **kwargs)
     42             return model.Var(name, dist, data, total_size)
     43         else:

~/anaconda/lib/python3.7/site-packages/pymc3/distributions/distribution.py in dist(cls, *args, **kwargs)
     50     def dist(cls, *args, **kwargs):
     51         dist = object.__new__(cls)
---> 52         dist.__init__(*args, **kwargs)
     53         return dist
     54 

~/anaconda/lib/python3.7/site-packages/pymc3/distributions/bound.py in __init__(self, distribution, lower, upper, transform, *args, **kwargs)
    153         super().__init__(
    154             distribution=distribution, lower=lower, upper=upper,
--> 155             transform=transform, default=default, *args, **kwargs)
    156 
    157 

~/anaconda/lib/python3.7/site-packages/pymc3/distributions/bound.py in __init__(self, distribution, lower, upper, default, *args, **kwargs)
     22             defaults = self._wrapped.defaults
     23             for name in defaults:
---> 24                 setattr(self, name, getattr(self._wrapped, name))
     25         else:
     26             defaults = ('_default',)

AttributeError: 'HalfNormal' object has no attribute 'median'

The following code works as expected:

with pm.Model() as model:
    D = pm.HalfNormal('D',1)
    S = pm.Bound(pm.HalfNormal, upper=D)('S', sd=1)

Please provide any additional information below.

It would be nice to have some consistency: if both D = pm.HalfNormal('D',1) and D = pm.HalfNormal('D',sd=1) work, then the variant with S = pm.Bound(pm.HalfNormal, upper=D)('S', 1) would work as well.

Versions and main components

• PyMC3 Version: 3.6
• Theano Version: 1.0.3
• Python Version: 3.7.2
• Operating system: archlinux
• How did you install PyMC3: pip with latest version on github (as of March 08)

[sshekh]

Hi @aakhmetz , I tried to look deeper into this. It's not just a problem with HalfNormal, but other bounded variables too. For example - the following example will work

with pm.Model() as model: 
     n = pm.Normal('n', 1.0, 1.0) 
     par1 = pm.Bound(pm.Normal, upper=n)('par1', mu=2.0, sd=1.0)

while the following will throw an error

with pm.Model() as model:
    n = pm.Normal('n', 1.0, 1.0)
    par1 = pm.Bound(pm.Normal, upper=n)('par1', 2.0, 1.0)

Bound calls _ContinuousBounded(name, self.distribution, self.lower, self.upper, *args, **kwargs) here which in turn calls Distribution.dist(*args, **kwargs). At this point the two sets of arguments (one to Bound: self.distribution, self.lower, self.upper and the other to Normal: - 2.0, 1.0 are mixed together in *args.

    def dist(cls, *args, **kwargs):
        dist = object.__new__(cls)
        dist.__init__(*args, **kwargs)
        return dist

At dist.init() thus:

When written as

pm.Bound(pm.Normal, upper=n)('par1', mu=2.0, sd=1.0)

args = (<class 'pymc3.distributions.continuous.Normal'>, None, n) and kwargs = {'mu': 2.0, 'sd': 1.0}
so it works as expected.

However, when written as

pm.Bound(pm.Normal, upper=n)('par1', 2.0, 1.0)

args = (<class 'pymc3.distributions.continuous.Normal'>, None, n, 2.0, 1.0) and kwargs = {}
So, Bound gets unexpected arguments while Normal gets none.

@twiecki , @ColCarroll, @canyon289 do you have any idea to proceed with this? I am trying to think of something which would stop the different sets of arguments from merging into one or always require parameter names when creating Bound objects.

I am new to pymc3 and trying to contribute thanks in advance 😅

[ColCarroll]

Wow -- this is tricky: thank you for the helpful writeup. I think I see what you're saying.

There is a lot of shuffling arguments around to internal functions -- do you think Bound, _ContinuousBounded and _Bounded can make sure that the arguments they use are keyword arguments? Specifically, replacing calls like

_ContinuousBounded(name, self.distribution, self.lower, self.upper, *args, **kwargs)

with something like

_ContinuousBounded(*args, name=name, dist=self.distribution, lower=self.lower, upper=self.upper,  **kwargs)

That seems like it might be straightforward, and not change any of the public API, right?

[sshekh]

@ColCarroll thanks for the response! I will look into it and update soon 🔜

[sshekh]

@ColCarroll I initially tried to convert all arguments to keyword arguments but dist.init will put the positional arguments *args (meant for Bound.distribution) before the keyword arguments no matter what, thus mixing them. So I converted all calls to _ContinuousBounded to use only (and all) positional arguments. PR #3446

[lucianopaz]

Closed by #3446