Blackjack with Q-learning

I use the python3 to create the blackjack with Q-Learning

In this blackjack, I prepared four decks for the dealer to use. If one of the deck is less then 30 cards, it will reset and re-shuffle.

The Q-Learning algorithm only set for the player, and dealer only follow the game rules (the dealer will continue hit the card if the cards score is less than 17.)

The Q-Learning algorithm will train a dictionary that key is the player's total cards score with dealer one card(which shows to the player) scores, and values are the numbers of wins.

This is the graph for different train/test times and winning percent result.