Effects of Population Size and Price Discretization on Outcome

With 6 prices, MY end up choosing 1 or 0.6, instead of undercutting each other. The MY still do better than the GT as long as their are not too many MY pricebots. It there are too many MY then the prices are usually 0.6 and the average prices and profits are dragged down. In order to have the MY undercut each other the number of prices must be increased. So, increasing the number of MY pricebots, regardless of the proportions of the other pricebots in the population, lowers the average prices of the MY pricebots, apparently asymptotically to 0.6, the lowest price that the MY pricebots will choose when there are only 6 prices.

It is interesting that in other situations with other starting situations, such as wA=0.25, wB=0.75, 6 prices, 100 pricebots and 100,000 rounds per generation, the MY do worse than the GT again, and the QL do much better than than the MY or df, almost as well as the df. For instance, after 150 rounds the pricebot proportions appear to have stabilized around 49 gt, 3 my, 0 df, and 48 ql, having started with 25 pricebots for each algorithm type. It may be that it is easier for it to learn a constant strategy like GT than a more changeable strategy like my.