Recreating DeepMind's AlphaZero - AI Plays Connect 4 - Part 1: Building the Game Environment

In this series we re-create DeepMind's AlphaZero to create an Artificial Intelligence Reinforcement Learning algorithm that plays Connect 4 using Python.
View the rest of the series: • AlphaConnect
Full Code: github.com/The...
AI, Deep Learning, Reinforcement Learning, DeepMind, Convolutional Neural Networks, Monte Carlo Tree Search, AlphaGo, AlphaZero, Connect4
pygame code developed from source: github.com/mah...

[@mathmo]

Nice! I think you can simplify the code a little though. You can get the valid moves as just
board.all(axis=0).astype(int) (or just leave out the astype(int), you probably won't need ints instead of bools anyway...)
with
and is_board_full as
board.all()
For checking the columns and rows in is_win you can use slices instead of the four equalities, e.g. to check the columns inside the double for loop you can just write
board[column,range(row, row+4)] == player
but here you can even remove the outer loop over the columns since you can do something like
(board[:,range(row, row+4)] == player).all(axis=0).any()
this way you can get rid of at least some of the code duplication. Also instead of checking the rows, columns, and both diagonal directions, you could get by having a nested helper function that checks only the columns and one of the diagonal directions and call it twice, once on board and once on board.transpose().
A good question to ask oneself would be how the code would generalize to accept not just more general board sizes, but an integer n instead of 4, and maybe even a board of dimension m.

[@discordraburg9988]

whats happening, isnt it the 3rd upload now?

[@theprincipalcomponent]

just the 2nd, realized there was a clip out of order and had to delete and re-upload it correctly. same content though, sorry for the confusion :)