Initial strats improv

[annadhm]

I thought about some imrovements for #8.
This code is based on @juwei95 code and also includes the #11 implementation

And added the following:

Currently, the walls of enemy paths are permanently added to ourMap and never removed. This significantly corrupts the map over time. So I implemented the removement of the paths from ourMap with every new gold pot.

For Path prediction of opponent players, I implemented a speed calculation. Making sure that our bot dont go for the gold if it would be to slow.

[juwei95]

Remarks

Thanks for fixing the map corruption issue.
I thought calling update_map every move would sufficiently clear the predicted paths of other players, but I didn't test it very thoroughly.

I evaluated your behavioral changes with my test script (#10), comparing the new code against the previous version (de6e8ea), similar to what @the-other-thanos suggested:

python3 monte_carlo.py -n 500 -r 100 -c 32

Evaluation

Here are the results across 500 games with 100 rounds each:

• With the low gold guard only, the bot performs somewhat worse (28% fewer wins)
• With the dynamic movement range prediction only, the bot performs significantly worse (72% fewer wins)
• With both new features combined the bot performs somewhat worse (36% fewer wins)
• The difference in performance grows for longer games over more rounds.

Conclusions

• Dynamically predicting the number of moves other players will take, based on the maximum they can afford, seems too pessimistic. In practice a small hardcoded value works better.
• The heuristic for the low gold guard does not appear to increase the efficiency of the strategy.
• Separating the objective map data from subjective path predictions is definetly a good fix and makes the code clearer 👍.

[annadhm]

Yeah true! I Would suggest that we push your #8 bot as our submission for Round 2, and only include the map seperation.