Week 8

This is amazing! 

This is so cool!

Can I ask how you did the actual calculation? I find that it's really difficult to calculate BECAUSE there are so many possible permutations. I have a similar calculation I did in Elite, leading up to Olympic selection, to identify the highest scoring team - where I run a script to calculate the team score based on each possible permutation, and see which team gives the highest score. But it only manages to run because there's a much smaller amount of gymnasts competing! I've given up on doing that calculation for fantasy because there are SO MANY gymnasts 🤣 

I do just do a brute force knapsack analysis (that is, compare every possible permutation). Given that the prices are categorical, you can prune your analysis set to the top 5 scorers per apparatus, per price, so it doesn't end up being computationally crazy. It's certainly not immediate, but still only takes about 10 seconds to run. I use the same analysis to draft my teams (after I do a bit of score balancing to account for recency and frequency factors), with extremely mid results 🙃!

Improvement to Week 2 (Same cost, max score of 396.675)

Replace Ella Zirbes on UB (cost 7500, scored 9.95) with either Utah's Abbi Ryssman or Utah's Bailey Stroud (cost 1500, scored 9.925)

Then replace either Isabella Trostel or Gianna Gerdes on BB (cost 1500, scored 9.925) with Florida's Selena Harris-Miranda (cost 7500, scored 9.975)

@gaia-steinberg 

At least the way I did it, the trick is to first cut down on the number of athletes we're considering before doing the permutations.

To start, for athletes that have a given price, I only need to consider those with the top 5 scores, since athletes with lower scores definitely won't be part of the highest-scoring team.

My algorithm goes roughly as follows (I wrote code in Google Colab that does this for me):

1. Take the top 5 scores for each apparatus for each price point (just realized that this step may actually be extraneous considering step 3 below, it's worth testing)

2. For each apparatus, get all possible combinations of prices for that apparatus (5 prices with replacement). If the sum of the prices is higher than our budget, discard it.

3. For each price combination, take the highest scoring athletes on the apparatus at those prices (for example, for price combination 7500, 7500, 1500, 1500, 1500, I'd take the top 2 scorers at 7500 and then the top 3 at 1500).

4. For each athlete combination from step 3, add up the scores to get the total score for that apparatus, and add up the prices to get the total price for that apparatus. Filter out score/price tuples where the price is higher but the score is the same or lower as another athlete combination.

5. Once we have all of the score/price tuples for each apparatus, find the combination of scores/prices across apparatuses that give the highest total score, filtering out those which are over budget.

I'd be curious to hear if @flippeur 's method is the same.

@rubyq1213 I wonder why my model didn't find that? A puzzle to work on, merci!