EA FC Automated SBC Solving ⚽

How does this work? 🚂

This project utilizes integer programming to solve squad building challenges (SBCs). The optimization problem is solved using Google CP-SAT solver. This approach offers extensive customization capabilities, alongside the ability to determine solution optimality.

The goal is to obtain the squad with the minimum total cost.

How to use it? 🔧

• To download the club dataset, use the extension (version >= 1.1.0.3).

• The inputs to the different constraints can be found in the input.py. Configure the appropriate inputs for each SBC constraint in input.py (L43-77 and L28-30) and then navigate to optimize.py (L581-615) and uncomment the relevant line based on the SBC requirements. Also don't forget to set the formation in input.py!

• For example, if the requirement is Same League Count: Max 5 or Max 5 Players from the Same League then set MAX_NUM_LEAGUE = 5 (L53 in input.py) and then uncomment model = create_max_league_constraint(df, model, player, map_idx, players_grouped, num_cnts) (L589 in optimize.py).

• If the requirement is Nations: Max 2 then set NUM_UNIQUE_COUNTRY = [2, "Max"] (L62 in input.py) and then uncomment model = create_unique_country_constraint(df, model, player, map_idx, players_grouped, num_cnts) (L598 in optimize.py).

• If you are prioritizing duplicates by setting (L28-L30) in input.py then model = prioritize_duplicates(df, model, player) in optimize.py (L615) should be uncommented.

• If for instance, the SBC wants at least 3 players from Real Madrid and Arsenal combined and at least 2 players from Bayern Munich, then set CLUB = [["Real Madrid", "Arsenal"], ["FC Bayern"]] and NUM_CLUB = [3, 2] (L43-44 in input.py) and then uncomment model = create_club_constraint(df, model, player, map_idx, players_grouped, num_cnts) (L581 in optimize.py).

• If the SBC requires at least 6 Rare and 8 Gold then set RARITY_2 = ["Rare", "Gold"]and NUM_RARITY_2 = [6, 8] in input.py (L71-72) and then uncomment model = create_rarity_2_constraint(df, model, player, map_idx, players_grouped, num_cnts) (L603 in optimize.py).

• Constraints such as Chemistry (optimize.py, L620) or FIX_PLAYERS (optimize.py, L623) do not require explicit activation. If there is no need for Chemistry, set it to 0 in input.py (L79). Similarly, if no players need fixing, leave FIX_PLAYERS empty in input.py (L12).

• The objective is set in optimize.py (L626). The nature of the objective can be changed in input.py (L20-21). Currently the objective is to minimize the total cost.

• Additional parameters in input.py should be reviewed for more information.

• In main.py, specify the name of the club dataset in L57. The dataset is preprocessed in preprocess_data_2 within main.py. Additional filters can be added in a manner similar to the existing ones.

• Currently the inputs are set to solve this SBC challenge. The final list of players is written into the file output.xlsx. To execute the program, simply run py main.py after installing the required dependencies. Note: This seems to be a very hard SBC and so had to enable the filter on rating in main.py (L41).

Dependencies 🖥️

Run pip3 install -r requirements.txt to install the required dependencies.

• Google OR-Tools v9.8

• Python 3.9

• pandas and openpyxl

Other Interesting Open-Source SBC Solvers ⚙️

• https://github.com/ThomasSteere/Auto-SBC

• https://github.com/bartlomiej-niemiec/eafc-sbc-solver

• https://github.com/kosciukiewicz/sbc-solver

Acknowledgement 🙏

Thank you GregoryWullimann for making the model creation process insanely faster.

Thank you Jacobi from EasySBC for helping with the squad_rating_constraint.

Thank you GeekFiro for testing the solver and providing valuable feedback and discussions.

Thank you fifagamer#1515 and Frederik for providing your club datasets (Real_Madrid_FC_24.csv and Frederik FC_24.csv) and your feedback.

Thank you ckalgos for creating the extension to download club dataset.

Thank you drRobertDev for providing the dataset Fc25Players.csv.

Thank you to everyone who have opened issues on the repo and provided their feedback.

Thank you to all the folks who commented on the reddit post.

Thank you FutDB for providing the API which allowed me to test the solver originally on 10k players (input.csv)