Squad Selection Issues: Part 1
Squad Selection issues: Part 1
Another year, another diastrous SC squad selection.
This year though I have decided to take a deep dive into my squad selection algorithm to disect why it so crap and to try make it better.
There are two main parts to the algorithm
- Part 1: Picking Cheapies
- Part 2: Picking Guns
To start I will focus on the picking guns side of the algorithm, which is more interesting. Picking cheapies is essentially picking rookies and cheap players that are playing round 1 (will visit this later).
After running part 1 of the algorithm the following cheapies have been added to the squad.
The total cost of the selected players is $3,504,300
I now have $7,945,700 left and 10 squad positions to fill with gun players. I want the players who will give us the highest scores week in week out.
To achieve this I have an algorithm that will maximise value of leftover players within the remaining budget. You can read more about that here ******. How do I judge a player’s value? I’ll get tot that later
Running this algorithm for the remaining 10 players with a budget of $7,945,700 we get
You will notice at a glance, that some positions will become full and I will not be able to add all these players to the squad. But I just go ahead anyway and start adding these players.
I was succesfully able to add the first 6 players, until I reached Alex johnson who could not be added as CTW was full and a positional swap wasn’t on the cards either. Read about positonal swaps here *******
So now my squad currently looks as follows
So I now need 3 front row forwards and 1 hooker. But I just rerun the algorithm for 4 players with a new budget of $3,226,000.
This will no doubt pick players that aren’t in these positions but I don’t want to remove these players altogether as a positonal swap may be available in some cases. If a player cannot be added, I then remove them from the player pool to avoid an infinite loop.
Running again, now with budget of $1,825,500 and players needed = 2
Now we need 1 player with a budget of $1,034,800
This give me my complete squad
So success! The algorithm has picked a squad.
Is it any good? I think most SC players would say… no, it’s terrible!
Although many of the gun players are seemingly top players. I can’t help but feel that there were better choices along the way. This is especially true as we ended up $330,300 under budget. Where is Tedesco, Cleary and the like?
The reason for this is likely because of how I have chosen to select a value for each player. At present, this is done by getting the total points scored for the last season.
This is clearly a pretty naive strategy. It would only take a player to have a rare spell on the sidelines for them to be severly undervalued. But in the other case you could have a player that will score highly in games but is injury prone and spends a lot of time on the sidelines. In which cases this should impact their value. I think I will next look into trying to train an neural net (AI) to try to predict how many points a player will score for the upcoming season. This can be used with the current selection algorithm to hopefully select better gun players.
Other things I noticed while running through this was that we ended up with 2 cheapie halfbacks because of a positional swap. This isn’t ideal, we definitely need to maintain at least one gun player in specialist positions (HOK, HFB, 5/8, FLB). Also I am probably selecting too many cheapies, this could be contributing to the squad coming well under the budget.