NotImplementedError: mean is not implemented: Categorical

[nbro]

I get the following error

NotImplementedError: mean is not implemented: Categorical

if I execute the following code

import tensorflow_probability as tfp

prob_dist = tfp.distributions.Categorical(probs=[1.0])
print(prob_dist.mean())

I get a similar error, if I attempt to call stddev(). I am using TFP 0.8.0 with TF 2.

The parameters mean and standard deviation are not usually defined for a categorical distribution and this error is due to the fact that tfp.distributions.Categorical inherits from tfp.distributions.Distribution, which defines the mean and stddev methods. PyCharm even suggests the auto-completion of these two methods for the tfp.distributions.Categorical, but this is at least misleading. Maybe tfp.distributions.Categorical should not inherit fromtfp.distributions.Distribution, which should not have those methods in the first place, because not all distributions have a well-defined mean or standard deviation.

[tomatitito]

I get the same error when calling mode() with an instance of the Multinomial or DirichletMultinomial distribution.

[rajpratyush]

@srvasude is this issue still open for work?