Modern strategy games with unique rules and frequent chance events are often hard to model.
The model is established based on Information Set Monte Carlo Tree Search (ISMCTS) and Residual Neural Network.
Graphical User Interface was designed to visualize the game process against AI over Fireplace simulator.
We compare the performance of our model under different conditions. Experimental results indicate that our model with ISMCTS rollout can provide move selection with certain possibility to win the game.