lambdaX | Partial evaluation for Python with syntactic sugar

You are not initializing your distribution; it's just how SciPy's distribution infrastructure is built. You are confused between an instance of the rv_continuous class and the rv_frozen class.

When you call your distribution class (custom_lgamma in your case), SciPy generates documentation and code for the user interface which creates the distributions we see and use from scipy.stats (e.g. norm, gamma) (this is what happens when you do custom_lgamma()). This instance contains methods such as pdf, cdf, etc that are called with the shape parameters (just like other scipy.stats distributions).

Calling this instance again (e.g. norm()) will create a frozen distribution (an instance of the rv_frozen class) whose shape parameters cannot be changed (which is what would happen if you call your custom_lgamma instance again or do custom_lgamma()()). This instance has the pdf method that just accepts the quantile parameter.

Here is the convention that SciPy uses:

It seems its a warning from the xgboost package. If you want to suppress you might want to consider something like:

Your code is misusing the AWS SDK. Specifically, it is using both the callback and promise features simultaneously. It should use one or the other, but not both.

I would change it to use just the promise (and obviously add some error handling to this):