Policy for testing types of input parameters

[lionelkusch]

Some functions have testes for checking the type of input.
For example: random_state in knockoff_aggregation in hidimstat/knockoff_aggregation.py.
random_state is a variable existing in multiple methods but this one is the only one which tests the type.

Which policy do we choose for testing the type of variable?

• Do we test all the input variables? (This can create a bit of slowing down the code.)
• Do we test only the most important?
  • if yes, how do we choose the most important variables?
• We don't test their type only to make the difference is their set or not.

[bthirion]

Testing a random generatoris a good idea. I would not (for the moment) test all inputs.

[lionelkusch]

Regarding some remarks in the PR #152, I think that we need to have a policy on which parameters should be tested.
Moreover, there are 2 ways to test the parameters:

1. with a function like the random_state, in the same way that scikit-learn
2. with assertion and an error message, like in the PR ADA-SVR (4/4): adding type and assertion #102.

[lionelkusch]

Another way can be to have one decorator for doing this check.
We can use as an example the one in scitkit-learn