AutoFlood_Demo1.ipynb

Hi

I want to offer my help on your venture of definitively solving FloodGame.

I can generate data for your use in learning and assessing your machine learning ai. - just tell me how you need it stored to be used seamlessly in your project.

[The Abridged Version Below] - Unabridged here

Since I saw your youtube video , I've spent my time addressing the same issue of what you thought about
I have a perfect foundational idea (in c++)

SUMMARY:

Deviation from your video:

• Does not yet use graphs*, doesn't plan to go to center.

PLAY FOR GAIN:

• I play the first option from the list of options with the highest tile gain in that step, (if playing color 1 or 6 gives the same expansion, play 1)
• I continue to iterate this way until I have win condition
• an integer containing the number of steps to that point is then set.

BRUTE FORCE:

• I use the integer (above- PFG) to limit the amount of steps ( the depth of recursion in essence)
• I then iterate into every first option of all choices, creating branches as I go, never playing further than the quick solution
• I then do backtracking and iterate through all (possible) plays - sometimes the order of the last 3 or 4 moves doesn't even matter, creating many paths, unnecessarily, but still necessary in the outlier cases where it actually does matter**(#1, 2 & 3) below
• in every iteration if a shorter path is found, set the integer to that and limit the amount of steps to the new, smaller number.

NEEDED ACTIONS:

1. I need to change existing code iterate BRUTEFORCE width first
2. I could rack an integer of the available colors still on the board (cannot win in 3 if there is 4 colors left)
3. reimplement a tracking plan to keep at least 'a' shortest path from the algorithm (removed because bad scaling)
4. processing efficiency - I know how I can get performance gain (if necessary.)

Since You'd want to pit your AI against the most efficient algorithm out there, I would say it would be my honor to create that algorithm to race you - or you could use the existing program(eventually) to generate data useable in your training of ml/ai.

If you would have me as assistant , and maybe give me some pointers along the way, I would like to help you on your quest to solve FloodGame - I would require you to tell me how to save my data, and what data you'd need to use for your purposes

Practicality: My 'play for gain' solution can generate both the grid and provide a reasonable solution in any configuration (ie: grid of 50x50 with 12 colors ) in a fast time. Though currently it sometimes gives an unrealistic value (my human brain least possible moves to win -1).
Potentially it will give a benchmark when you're planning to slap a neural network on it, (PLUS the flexibility of the width/heigh/depth dynamic might be an interesting 'curveball' to validate your training data and neural network).

Sincerely

LeAnnAlexandra