DSL for defining markets for sportsbetting.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


Flawed prototype of a DSL for defining markets at Geneity.

2014's Christmas project.

See http://www.martin-griffin.com/blog/geneity.html#market-dsl


Hi all,

I finally got around to creating some proof of concept code for
representing markets as functions of the state of an event (realized as
an expression that transforms a property of the state).  It's looking
very promising, I have a pretty clear idea of how to evolve the design
and implement the features I want.

Here's an example of some of the stuff it can do already (based on

# No goals.
>>> football0 = football(['home', 'away'], [])

# Home has scored one goal.
>>> football1 = football(['home', 'away'], [period.begin, goal('home')])

# MRES.  The team(s) that has scored the max number of goals.
>>> most_goals = max(map(partition(goals, 'team'), len))

# A tie is represented as both teams, so we can support n-way sports.
>>> result(most_goals, football0)
set(['home', 'away'])

>>> result(most_goals, football1)

# How we expect the prices to shift between football0 and football1.
# TODO: The keys of this dict should be sets like for result.
>>> trend(most_goals, football0, football1)
{('home', 'away'): -1, ('away',): -1, ('home',): 1}

# There's a static grid in gfootball.py that favors away.
# TODO: The keys of this dict should be sets like for result.
>>> calculate = gfootball(football0)
>>> calculate(most_goals)
{('home', 'away'): 0.3 ('away',): 0.6 ('home',): 0.1}

example.py also contains TOTG and something similar to FGSC but with
teams as selections.

The code itself is... okay.  It's very much a work in progress and many
things that you would expect to work will blow up with an AssertionError
(if you're lucky!)  Some of the algorithm code I know is wrong and yet
more of it is likely wrong without me having realized yet.  The next
things I'll be working on are, in roughly priority order:

- Support attr and nth in trend.

- Choose a better (and consistent!) set of names for the modules,
functions and variables.

- Add tests that verify eval, trend and calculate are in-line with each
other; e.g. if eval(expr, state) has a result then
calculate(state)(expr) should return 1 for that result and 0 for the others.

- Make gfootball look at the state when creating its grid; currently you
get some very interesting results when asking for the probabilities of
an event that has goals.

- Make eval fail if the periods it is looking at are open.

- Extend typeof so that it returns more useful, non-adhoc types; rewrite
members so that it uses those types.

- Support market definitions with free variables (e.g.
most_goals_in_period where which period isn't specified in the
definition); this is equivalent to market extras.

- Reorder parameters to eval etc so that the most likely to be fixed
parameters are on the left (makes partial application possible); this
requires using something other than singledispatch.

- Add automatic dispatching on properties of the operations that
couldn't reasonably have identical implementations, e.g., the function
to map or the collection to retrieve.

- Infer which markets could be handicapped, over-under-ed and
odd-even-ed (are there more?).

- Infer the data currently in tMktSortSport.

- Overhaul state; it's a quick hack currently, and I think I have a much
better design that will lend itself to a similar level of introspection
(e.g. to automatically build the in-play UIs).

I'm pretty sure that, if we wanted to, we could incorporate this code
into the legacy sportsbook without *too* much effort.  Let me know if
you think this is worth pursuing and I'll try to come up with a
migration plan as I work.

As usual your feedback is greatly appreciated,