# Progress: New evaluation idea

While playing a bit of drunken Tetris, I came across an idea that I use when playing: probabilities.

By evaluating a field based on how each different piece can fit in the field, one can get the best possible field. For instance, A human player often refrains from creating a chasm because there is only 1 specific piece that can fill it. However, they will be more likely to make a risky move and create a 2-deep well because 3/7 pieces can fill that well. Similarly (I think), creating a contour {1, …} (from the left wall) is worse than making a contour {-1, …} because it may be able to fit more pieces.

By making an evaluation function that evaluates the field based on how the pieces can fit on it, the strategy may improve. This strategy will also be quite useful for 2-piece Tetris (I think, drunkenness may be affecting my thought processes) because the agent can set-up the field for the next piece by maximising the next piece’s probability of fitting.

Edit: Something I forgot to mention was weighting the probabilities based on the observed piece distribution as well. There would be little point creating gaps for pieces that are unlikely to occur. So adding into the field evaluation calculations are the piece multipliers.