Predict Output Screen
Using Predict
To evaluate how good your k_factor is, look at the percentages in the output table. You want to minimize the error percentages by adjusting the k_factor. If you can get low errors, which I can, then that directly affects the success percentage and therefore the parlay recommendation. Setting the parlay threshold to a reasonable value like 1.05 or 1.1 will cause parlay recommendations when the "edge" is above that threshold value. You can then multiply the edge numbers of the parlay you are constructing to see the "value" of the parlay.
Next we consider the prob_multiplier. You want to set this value according to the average point spread of the sport, so high for NFL and low for MLB as examples. It should be in the neighborhood of the point spread that would occur if there were a nearly 100% chance that the favorite would win. Good values to try range from 5 to 40 depending on the sport, and any positive value will work. Sometimes really large unrealistic values work well too.
To optimize the point spread we set the point_spread_div, point_spread_pow, and point_spread margin values. The div and pow values affect the formula: ruts = (ps/div)^pow, which is the running-up-the-score contributionn to the point spread estimate. It models the behavior where favorites run up the score against underdogs. Reasonable values for div are 50 to 250 and pow are 0 to 3. set pow to 0 to turn off the rutsing feature. The point_spread_margin defines a "middle" around the point spread odds where the program will not place a wager. It is kind of like an extra safety zone and larger values drive up the success rate while smaller values provide more recommended bet games.
The over/under div, pow, and margin work the same way as their point spread counterparts. The over/under avg controls how slowly or quickly the algorithm adjusts to new average points scored values that occur. Try 0.8 for a very rapidly tracking algorithm that forgets quickly and learns quicly, and 0.99 for a slowly adapting algorithm that stays fairly constant over time.