This is a great question. and I am sure many people have a view on this which would be good to see.
I think this is one of the challenges of the field;
There are many data science notebooks that provide ways to implement many of the existing algorithms, but I believe these are limited in terms of being able to modify any of the lower level aspects of the learning algorithm.
There are simulators that are more useful for research that allows more tailoring of the algorithms.
Plus there are alot of open source code that you could use as a starting point.
Take a look at http://homeAI.info as a starting point for information on tools and software
Also if you are interested in a specific research paper, you can also contact the authors and find out what tools they used.