Fotball season simulator

Simulate the remaining matches of the season. This program computes every possible configuration of outcomes in the remaining matches by brute force.

Currently, this is limited to a tournament with 14 teams, and the input data is gathered from a .csv file of the season fixtures. As an example, look at fotball.no.

You can specify which rounds you want to simulate, and all unplayed games in these rounds will be run. Because of the combinatorical explosion, anything above three rounds is infeasibly expensive to run. The number of simulated results is depending on the number of outcomes in each game (home win, away win, and draw), and the number of games simulated:

num_outcomes_total = num_outcomes_per_game ^ num_games

Running on a computer with 12 Intel(R) Core(TM) i7-3930K CPU @ 3.20GHz processors, and 32 GB memory, the following is approximate timings based on one run.

Number of games	Total number of outcomes	Time (s)
7	2 187	0.0077
14	4 782 969	12.41
21	10 460 353 203	28 627.96

Build

If you do not have rust installed, you can install it by following the instructions in https://www.rust-lang.org/en-US/install.html.

Then, you clone this repo, and build it as

cargo build --release

You need the fixtures, which can be downloaded from https://www.fotball.no/fotballdata/turnering/terminliste/?fiksId=158515, note that this is in Excel format, and needs to be converted to .csv.

Run

When this is done, you can run it as

./target/release/football_simulator --season_fixtures /path/to/fixtures.csv

Note that currently, which rounds are simulated is hard-coded. If you want to change it, you need to change the rounds variable in src/main.rs. Once it is changed, you need to recompile the program.

This program can be used to compute two different things. One computes and stores every possible table outcome, and this is invoked by the --task=tables flag. The other one computes distributions of the possible positions for each team, and this is invoked by the --task=standings flag. Note that the table version is very memory expencive, and I have not tested it above games. The other version does not require a lot of memory (as the tables are not stored), but is a bit slower since it does every computation as the table option, in addition to some other computations (the timings above are for the --task=tables version).

Plot distributions

You can use the plot_standing_distributions.py to generate plots of the distributions

python plot_standing_distributions.py -i=/path/to/outcome_distribution.csv -o=/result/directory

Example

Results after simulation of round 24, 25, and 26.

Team	1	2	3	4	5	6	7	8	9	10	11	12	13	14
Ready	0	0	0	60859107	208635426	429284043	897513453	1463553522	2320194861	4274471088	791492796	14348907	0	0
Follo	0	0	0	551124000	946636389	1312880157	1691379873	2140478136	2066920578	1631497626	119436444	0	0	0
Rilindja	0	0	0	0	0	9639	6076863	57634902	226673640	784991655	5596146468	3214863756	573956280	0
KFUM 2	0	0	0	0	0	0	0	5511240	56805138	209446803	3278794140	5659846650	1249949232	0
Kristiansund 2	0	0	0	0	0	0	0	0	0	0	256686003	1567219509	8507307528	129140163
Drøbak-Frogn	0	0	0	3627800946	2305730043	1800739350	1288291230	930826566	423356544	83608524	0	0	0	0
Hødd 2	0	0	0	0	0	0	0	0	0	0	0	0	129140163	10331213040
Oppsal	10446004296	14348907	0	0	0	0	0	0	0	0	0	0	0	0
Træff	0	258280326	10202072877	0	0	0	0	0	0	0	0	0	0	0
Spjelkavik	0	0	0	2637643500	2722007268	2043194544	1427788539	958640103	531571491	136737558	2770200	0	0	0
Lokomotiv Oslo	0	0	0	2407779594	2180030031	2120798781	1675158651	1154494701	723324735	195413310	3353400	0	0	0
Molde 2	14348907	10187723970	258280326	0	0	0	0	0	0	0	0	0	0	0
Ullern	0	0	0	396827991	871826409	1241946108	1556164062	1856737323	2245996674	1955955519	330824736	4074381	0	0
Herd	0	0	0	778318065	1225487637	1511500581	1917980532	1892476710	1865509542	1188231120	80849016	0	0	0