Refactor Continuous Distribution and Continuous Factor

[ankurankan]

• New BaseDistribution Class
• New CustomDistribution Class
• Refactor old ContinuousFactor Class

[coveralls]

Coverage decreased (-0.9%) to 95.32% when pulling 6f45aa5 on ankurankan:issue/782 into 5d885cb on pgmpy:dev.

[coveralls]

Coverage decreased (-0.8%) to 95.432% when pulling b791b76 on ankurankan:issue/782 into 5d885cb on pgmpy:dev.

[yashu-seth]

@ankurankan I think methods like reduce, marginalize etc. should be in the base distribution not in the custom distribution. There can be cases where I have a different (better) implementation for a particular method but I want the other methods to be inherited.

For example, if we define a gamma distribution and we know the direct results for reducing a multivariate gamma distribution we can implement it by overriding it. But for its other methods, for example product we want the default behavior where the PDFs get multiplied, then we would want to inherit it from the base distribution.

[ankurankan]

@yashu-seth Yes you are right. Also would you like to take over this from here ? I will be quite busy with my exams for the next couple of weeks.

[yashu-seth]

@ankurankan Okay. I will look into it.