I’ve been getting my Zoltar NFL football prediction system ready for the upcoming 2017 NFL football season. Every year I tweak Zoltar with new algorithms and new features. This year, I decided to put in an explicit, natural language advice module.

In previous years, Zoltar would just spit out predictions, for example this dummy data:

Zoltar:    patriots  by    6  dog =      chiefs    Vegas:    patriots  by    2
Zoltar:       bills  by    6  dog =        jets    Vegas:       bills  by   11
Zoltar:      ravens  by    0  dog =     bengals    Vegas:     bengals  by    1
Zoltar:      texans  by    9  dog =     jaguars    Vegas:      texans  by  6.5
Zoltar:    dolphins  by    5  dog =  buccaneers    Vegas:  buccaneers  by    8
etc.

Then I’d have to go through the output manually and infer what Zoltar would suggest. For example, in the first game listed, Zoltar thinks the Patriots will win by 6 points but Vegas believes the Patriots will win by just 2.0 points (expressed as “Patriots -2”). If I set a threshold of 3.0 points, then Zoltar would recommend a bet on the Patriots, believing they’ll cover the spread (win by more than 2.0 points plus the threshhold).

In past years, every now and then I’d make a mistake in interpreting Zoltar’s output — it’s trickier than it might seem. So, this year I’m adding a module that explicitly gives (hypothetical only) betting advice. For the prediction data above, the advice module would display something like:

Zoltar's advice: (on dummy data!!)

chiefs         vs       patriots : bet Vegas favorite patriots (favorite will cover the spread)
jets           vs          bills : bet Vegas underdog jets (favorite won't cover the spread)
ravens         vs        bengals : Zoltar disagrees with Vegas on winner but advise no bet
jaguars        vs         texans : advise no bet
buccaneers     vs       dolphins : advise bet on underdog dolphins who will win outright

The logic is mildly tricky. For example, if both Zoltar and Vegas agree on who will win, then if the predicted margins of victory are close — within some threshold value (typically 3.0 points), then no bet is recommended. Else if Zoltar’s predicted margin of victory is greater than the Vegas predicted margin of victory, by more than the threshold, then the advice is to bet on the Vegas favorite because Zoltar believes the favorite will cover the spread. And so on.

I’m writing this post on Friday, Sept. 1, 2017 — I hope to get Zoltar’s first-week NFL predictions posted by this coming Tuesday, Sept. 5, 2017 for the first game of the season to be played on Thursday, Sept. 7.

A previous version of Zoltar with a speech interface