Recommendation system based on previously registered data

Question

I’m a machine-learning layman I started a course at Alura about classification system using Pip but it does not solve my problem, I would like a library for a recommendation system, i would collect data related to each match of an electronic game and recommend the best route for the player to follow.

The logic I develop, just would like some recommendation(A North) of libraries and/ or languages to develop this.

EDIT: I will apply machine-Learning in a game, at the end of each match the player would fill in dice like: where he landed on the map, how long he stayed alive, how many players he killed and with which weapons and especially in what position he was in at the start. With this I would train, analyze, or whatever an algorithm is to give him tips for example where he should fall, what weapons he should use and other things.

Answer 1

Are you sure your problem is one of recommendation? It fits into problems of recommendation those where the number of items that can be recommended is very large and so cannot be solved in a scalable way by common classification algorithms. In your case, it seems to me more a problem of multiple classes, in which the number of possible routes is not very large.

For recommendation, a common approach is to use algorithms based on the singular value decomposition of the interaction matrix. This package in R has a pretty cool implementation of one of these algorithms. Spark has an implementation of ALS which seems to be very scalable. In python, there are also several options. I quite like the Lightfm. But there is also the Surprise.

Otherwise, most classification algorithms also have extensions for multiple classes. Generally, the call is made One-vs-Rest, train an algorithm to classify each class versus the others and then choose the class with the highest score.