An extendable collection of ideas in order to improve the current state of the repository in order to create an improved user experience.

• All data for the examples might be generated on the fly?
• Wrapper should maintain all user specified methods and attributes. Implementing a hybrid approach where a) it could be inherited by it and b) taking an initialized class and copy it to a new class.
• Provide abstract method i.e. for reduction of TC in order to be implemented by user
• Up to second order: check if Hessian/Jacobian implementation is faster and switch to this, instead of custom multiple grad calls.
• Output in plain dictionaries: the keys might be the coefficients w.r.t. the derivative and values the corresponding calculated TC values
• Output should contain all information, off-diagonals (0, 1) <-> (1, 0) should be kept distinct, the decision of its summarization should be forwarded to the user. It might be useful to allow only a function to pass one/two/three numbers for first/second/third TC of given indices or a list of 1/2/3 lengths tuples.
• The current checkpoint implementation would not be necessary then given the point above and could be dropped. It would be required to store the returned TC calculation output each optimization step/how often they want it to have.
• Only a very basic plotting example (in README) should be given, without the commitment, since this would be very individual to each user
• More things to be considered?