Add mode method

[aloctavodia]

In #603 we introduced a mode argument to maxent and a mode method to many distributions. The mode is still missing from the following distributions.

• AsymmetricLaplace
• BetaScaled
• ChiSquared
• ExGaussian
• Kumaraswamy
• LogLogistic
• LogitNormal
• Pareto
• Rice
• SkewNormal
• SkewStudentT
• StudentT
• TruncatedNormal
• Uniform
• VonMises
• Wald
• Weibull
• BetaBinomial
• Binomial
• Categorical
• DiscreteUniform
• DiscreteWeibull
• HyperGeometric
• ZeroInflatedBinomial
• ZeroInflatedNegativeBinomial
• ZeroInflatedPoisson
• Dirichlet
• MultiVariateNormal
• Hurdle
• Censored
• Truncated
• Mixture

[rohanbabbar04]

I have a question: there are some distributions that are bimodal. Should we return both modes or just one?

[aloctavodia]

Good question. Maybe only the major mode? Not sure what to do if two equal modes.

Also for the beta and beta scaled, I think we are currently returning 0.5 when alpha and beta are lower than 1. But we should do:

0.5 if α=β = 1
No mode if α<1 OR β<1.
at 0 for α ≤ 1, and at 1 if β ≤ 1

[rohanbabbar04]

One option is to go with the major one, other is to return a tuple of both modes

[aloctavodia]

yep, probably the better thing is to respect multimodality. Then we will need to check that setting the mode works for maxent or that at least a warning is raised if we can not handle those cases.

[Advaitgaur004]

Have added

• StudentT
• Discreteweibull
• exgaussian
• hypergeometric
• logitnormal
• skew_student

[Advaitgaur004]

One option is to go with the major one, other is to return a tuple of both modes

I believe returning a list or tuple of modes is beneficial. For a bimodal normal mixture, both peaks should be returned. That’s my suggestion.