I have some amazing news.
FPL Optimized has been named a finalist for Best in Fantasy Football – Organisation at the Football Content Awards 2024! This wouldn't have been possible without all of you. Now, I need your help—every vote counts! 🙌
🗳️ Cast your vote here:
#FPL
If you had a lineup of Sá, Cancelo, Trippier, Schär, Saka, Trossard, Foden, Almirón, Toney, Haaland (C), Kane (V), and Begović, Reed, Perraud, Amartey in GW1 and forgot about your team, you would have accumulated 874 points by the end of GW12.
#1
OR had 824 points.
#FPL
Jokes aside, GWs like these remind us there is no way to control outcomes. No one is "genius" for bringing a player. It is all probability, you can get lucky. No one "can learn from mistakes" as there are no "mistakes". Planning helps, yes, but sometimes it is just luck.
#FPL
Here is the optimal squad with 50/50 objective weights to predicted points, and minimum loss against field (safe, high own%). Analytics with wisdom of the crowd mixed in.
GW1 prediction with 58 lineup points
#FPL
For risk-takers, here are top 10 players with highest expected gain over 10K average and predicted EO.
KDB +9.9 4.18%
Dias +9.6 0.73%
Mahrez +9.5 0.23%
Mané +9.4 0.82%
Ederson +9 3.05%
Tielemans +8.4 0.02%
Foden +8.3 10.27%
Alisson +8.3 0.35%
Barnes +8.2 6.48%
VVD +7.7 5.73%
For the 100th time:
Yes, I play FPL with modeling and data
Yes, I enjoy it
No, I don't feel like a robot
Yes, everyone can play any way they want
No, I know models have biases, but more so eye-test
No, it doesn't bother me
Yes, we can't plan everything
No, it's not pointless
#FPL
With
@FPL_Spaceman
👨🚀, we wrote a tool that can check how similar a player's decisions to the content creators. Now you can find team, transfer, copycat similarity of any player you want! It will be online in a bit.
#FPL
I have checked Mikkel Tokvam's FPL projections, and here is the optimal no-transfer squad for first 6 GWs.
(Fixture/xPts/xMin data in boxes)
Alisson
James, TAA, Perišić
Trossard, Salah, Foden, Elanga
Toney, Mitrović, Haaland
(Gazzaniga, Tierney, Jonny, Andreas)
#FPL
I've been learning about creating animations recently and prepared this video, highlighting:
🔸shots
🔸attempted assists
🔸expected point - FPL point comparison.
Let me know what you think!
FPL Analytics and Optimization extension is finally published in Chrome Store. 🥳
It provides shortcuts to Live GW Tracker and Simulated Scenarios pages directly from the FPL website.
#FPL
#FPLCommunity
Hey everyone, as we are moving into final GW, here is a page you might enjoy checking (I hope) if you want an analytical evaluation of your season:
Season Highlights!
#FPL
Top 3 James replacements by ensemble data EV (Defenders below 6M). The first viz shows best picks in specific horizons. The second viz shows the top picks for each GW in detail split by data source.
#FPL
Hello everyone.
As many of you are aware, there's been a recent stir in our FPL community regarding the leak of Kyle Walker not starting in today's game. I want to address this openly and transparently.
1/6
#FPL
Let's try building a few WC squads using optimization and, well, variance tapping. I will add random noise to MD points and generate top 3 WC squads. Comment with
- a random number between 1-1000
- your team ID
- up to 2 players you want
- whether you are using BB36 or not
#FPL
Hey all, I talk about optimization and optimal squads in FPL all the time, but you might be wondering the logic behind it. Here is a small thread about what is optimization, its history, when it is useful, and why we are applying it to FPL. 1/18
#FPL
Maddison YC ban is annoying. Here's top midfielders under 8.3M by projected points using different data sources until the World Cup.
Zaha, Bowen, and Foden are top three by total EV in average.
#FPL
7 GW no-transfer raw optimals from available sources.
Mixed data (with weights based on EV scale) gives a squad that performs better in average.
Ederson
Gabriel, Estupiñan, Stones
Saka, Mbeumo, Mitoma, Fernandes, Rashford
G.Jesus, Haaland
#FPL
It was optimal to have Kane in GW1, and to have Haaland in GW2. If you are evaluating your decisions by the outcome, you are not doing yourself a favor. Single player/GW outcomes are not important, consistent positive EV difference to field is.
#FPL
#FPLCommunity
I have checked the brand new prediction data on FFHub and found the optimal plan for the first 8 GWs. It is solved to optimality with a MIP model.
Starting lineup:
Leno
Dunk
Veltman
James
Shaw
Buendía
Salah (C)
Fernandes (V)
Sarr
Watkins
Calvert-Lewin
1/5
#FPL
🔥We have a new podcast for you: FPL Optimized!
Every week in 30 mins,
@BelfiBB
and I will delve into analytical side of FPL and bring you the state-of-the-art using eye-test and spreadsheets.
In our first episode, we discuss AI in FPL.
#FPL
For BB29, this is the "ultimate team" that maximizes analytical projections (FPLReview) with current minute predictions: 130.33 EV.
It might not be ideal for everyone, as you might lose EV in GWs before or after, but a good reference to have.
#FPL
#GW13
Most people brought Trossard as a short-term punt, here are alternatives (below 9M) until the World Cup using different data sources using early numbers. Most models favor Foden, except Fix (Martinelli) and Review MD (Saka) in total EV. Values are pretty close.
#FPL
A preview of top 5 players by projection models (expected value / EV) between GW17 and GW24 for each position.
Overall top 5 players by ensemble model (average of all models):
🔵66.48 Haaland
🔵53.86 KDB
🔴48.78 Salah
⚪️46.76 Kane
⚪️42.66 Son
#FPL
Your favorite party pooper fun fact is back with an update :)
If my math is correct, if you had this team in GW1 and completely forgot about it until now, you would be at 2595 points and 35th in overall rank.
Thanks to
@FPLFocal
for reminding me to run it again.
#FPL
If we knew how many points every player would score beforehand, the only way to win FPL would be using an optimization model and solver. There are too many combinations of decisions that you realistically cannot calculate it otherwise. 🧵
#FPL
A quick look at raw FH32 optimal teams from major FPL prediction data sources with current values and minute predictions. TAA and Salah are only players that appear in all of them. All of them includes triple Arsenal and Liverpool.
#FPL
I'm wondering if I should stop playing FPL myself, and instead focus on content generation and maybe working with a few interested people and do optimization for them. I am kind of growing tired of getting close decisions wrong, week after week. Need to think a bit.
#FPL
As promised in the pod, here is the raw optimal WC26 + BB29 + FH32 using
@fplreview
projections. It optimizes the period between GW26-34. No banned or forced in players, using default parameters, forced no GK rotation, 0.99 decay base.
(Sharing again since noticed a mistake)
No
#FPL
stream today, but let me give you a glimpse of my decision making process.
It's no secret that my play is based on expected values (EV) but here is how a typical GW goes.
#FPL
With current Review values, optimal attacking premiums (10M+) for first 6 GWs are Salah + Haaland. Here is how much worse other top 10 premium combinations are. Obj. score gives more value to closer GWs (decay 0.85).
Salah+Haaland / Salah+Kane / Salah only are good options.
I was checking optimal Free Hit squads for DGW22 (with random noise added with a factor of 30%) and noticed that 3 players are always in the optimal lineup after 100 iterations:
Ronaldo, Son, Alonso
Best captains: 54% Ronaldo, 31% Son, 12% Kane, 3% Wood
#FPL
I think I will slide down in ranks heavily once people start using their BB & WC & FH chips in following GWs, so saving this as a memory. My best OR, best percentile (<0.02%) thanks to my luck, and of course optimization modeling.
#orms
#FPL
For me, WC8 optimal team schedules 4 transfers
GW9 Son to Salah
GW11 Cancelo + Haaland to TAA + Kane
GW13 Kane to Haaland
WC9 optimal team schedules only 2 of these
GW11 Haaland to Kane
GW13 Kane to Haaland
If I value FTs at 1.8, WC9 is a step ahead.
For unknown reasons, someone has decided to make false statements about my treatment of people of color and women. I thought that ignoring the comments would exhaust the person/persons harassing me and eventually it would stop.
1/3
#FPL
Wrote an optimization model that maximizes how template a team can be. Using replies to Andy (video coming soon👀) I scanned teams with my scanner and then optimized it.
Here is the most template GW1 team from Twitter. (1.5 ITB)
How many players do you have in your draft?😅
#FPL
Feeling down due to your rank? Last season's winner was at 120K after 8 GWs (~84 points below
#1
) and average OR of the final top 150 after GW8 was 82K. Worst rank in this group was 800K, who eventually finished at 11.
Lots of things can happen...
FPL is a strange game.
#FPL
Using available FPL prediction data, I have generated a visual displaying top 5 players in each position and price category. Numbers represent their average EV (ensemble). Top 3 players in each category indicated by gold/silver/bronze medals.
#FPL
I will share optimal teams as prediction data comes back online from various sources.
The first data is from Drafthound based on bookie odds. Here is the 10-GW fixed bench raw optimal team.
(No transfer, no decay, fixed bench)
A light 🧵 on my observations regarding
#FPL
so far.
I enjoyed watching football since childhood but strangely never heard of FPL until the beginning of this season. One of my work colleagues that I do optimization projects with mentioned it, and I took an immediate interest.
#FPL
I hope everyone is enjoying their break from FPL 🌴
I know FPL chips are probably the last thing you want to hear on the beach... But in case you do, I crunched some numbers that you might enjoy before we decide to forget all about this season or FPL Towers wipes the data.
So far in
#FPL
I have collected 1487 points
534 from minutes
400 from goals
231 from assists
212 from clean sheets
...
On the other hand I lost
-10 from missed penalties 😑
-12 from hits
-19 from goals conceded
-36 from yellow cards 😓
#FPL
Early numbers as Salah alternatives. Salah still has solid predictions, but if you are looking for alternatives, here is what models are suggesting.
#FPL
#FPLCommunity
As we keep trying to build new tools and methods for FPL decision making, one thing that we hit our heads over and over is the randomness (aka variance). Existence of variance is often raise skepticism towards optimization efforts. 🧵
#FPL
The optimal team is relative to your parameter choices, and is a moving target... but with current values, here are 7-GW horizon, no-transfer optimals* by different data providers.
Under each plan (column), I have listed score evaluations by the other data providers.
#FPL
#FPLCommunity
Optimal 8 GW plan using Review's preliminary values. It will take a while until HiveMind minutes to get more accurate, but I dropped players with less than 55 exp. minutes.
Plan 1 gives 456.73 xP
Plan 2 gives 453.61 xP
A preliminary version of the Massive Data model for the 21/22 season is now live!
The official start of 21/22 for will be w/c 2nd August, some updates on work being carried out below
#FPL
#FPLCommunity
20K followers—wow! 🎉 This milestone is about all of us who believe that data and analytics can make fantasy sports even more exciting. Thank you for being part of this incredible community! Here's to the next 20K. 🙏
#FPL
2021-22 Season is over. What a season it has been! I finished my second season with 2663 points and at 3,481 OR. A tier of %0.17 if you assume only 2M people stayed active until the end.
My first season OR was 304,304 😅
#FPL
Hi all, I have updated my Python codes for FPL optimization. Here are steps to run it. (For existing users, we will move away from login method to team.json method)
#FPL
Here we go! GW1 squad.
After running the solver for lots of hours, this is the squad I have ended up with. Some structural constraints, some minutes adjustments, and we are finally ready.
I hope it will be another golden year for analytics. 🏆
Good luck everyone. 🔥
I wasn't sure what to expect at the beginning, but finished 1st season of
#FPL
with 2315 points, 304,304 overall rank (within 4%). Produced 7 tutorials on optimization, wrote 4 blog posts & built an open-source webpage with several tools. It was fun, let's try again next season!
I have joined Fantasy Premier League
@OfficialFPL
for the first time this season, using a multi-period Knapsack optimization formulation. My first week rank is 475K out of 6.1M players, so my team "Optimized FC" is in top 10%. I am using pure optimization output to pick my squad!
Testing a new
#FPL
visualization in Official FPL style.
Used Market Odds data from
@fplreview
and merged it with FPL's brand new xG (Opta) values. Last 5 games market goal predictions (adjusted by player mins), xG, and goals (excluding OG) for all PL teams.
ARS, AVL, BHA, BOU
So proud to be sharing this.
3 of 5 top
#FPL
mini leagues are analytics ones:
#1
is our podcast mini league (390 teams)
#2
is our analytics discord mini league (240)
#5
is our analytics twitter mini league (488)
---
LetsTalkFPL has 124,000+ teams
BlackBox has 15,000+ teams
#FPL
Hello all new and veteran FPL players 👋 We have been publishing a podcast (FPL Optimized) with my good friend Bas for a while. I'm a vocal advocate of the use of data and analytics in FPL. I have a chance to host a guest this week, while Bas is on the beach 🏖
Hi everyone, decided to create a 🧵to answer some questions I received for FPLOptimized.
I will go over briefly what each page/tool does, how to use them.
I ran an interesting experiment! If you are familiar with the "ghost racer" concept, I applied the same idea of replaying last season alongside top FPL managers to find out what they did different than optimization would have done. I'll share some of my findings soon.
#FPL
Analytical methods make it easier to measure the risk you are taking. Here is 6 GW no-transfer optimal using latest
@fplreview
data with alternative players listed below. The loss in expected points when picked is shown for 40 most owned players (excluding low xPts players)
Hey
#FPL
community! Join our Summer Python Camp in the FPL Analytics Discord! We're self-learning Python with a free e-book, covering 2 chapters/week, starting this Saturday (June 1st). Details and sign-up in our server. Link in next tweet!
#FPL
I would like to say a few things on this for the record, because we discuss similar topics periodically.
I'm sad to hear Mark's experience on this, but prediction models are not certain, there is always room for error. Think them like weather forecasts.
I love the use of models for FPL. I was a part of the team that delivered one of the first (RMT). What I don't like is how they appear to have established what's perceived as the "right" or "correct" or "optimal" solution for every decision we now make...
#FPL
Doesn't matter, but finished 61,695 (just above Andy Martin 😎).
My highest OR this season. Quite a drop from last year (3,481) but it was boiled down to few split decisions. It was an exhausting season. Thanks everyone and especially FPL analytics family for keeping it fun.
"Is it fair for fantasy football managers to rely on AI?"
Yes. It is more of a strategy game than a football game. AI (ML) part is only relevant at prediction aspect. Football has enough variance to keep game interesting even when playing with data.
#FPL
Optimal WC10 depends on your team value, parameters, and data source you use. Here are some optimal WC10 squads* from various data sources.
(Will share optimal teams for lower team values below)
*7GW horizon, 101.7M team value, linear bench weights, max 2 sch. transfers
It might be too early for some to start thinking about
#FPL
but here is a 🧵on optimal team rotations.
I have used
@BenCrellin
's spreadsheet to read fixtures and FiveThirtyEight's club ratings (SPI) as of now. I will update it when we get closer to the season.
#FPL
If you changed for FPL team this season only twice (GW1, and GW17 after World Cup, no changes in any other GWs) as
GW1-16
Lineup:
Henderson
Gabriel, Saliba, Trippier
Trossard, Rodrigo, De Bruyne, Almirón (V)
Toney, Haaland (C), Kane
Bench:
Ward, Mykolenko, Reed, Perraud
#FPL
Used Mikkel Tokvam's data for a sensitivity analysis (40 iterations). There are 6 players who appear in optimal GW1 team more than 70% of runs:
- Fernandes 100%
- Zinchenko 97.5%
- Mitoma 95%
- Gabriel 92.5%
- Rashford 87.5%
- Haaland 85%
#FPL
Review's model is about to arrive. Is there anything you say you wouldn't follow even if it is optimal?
5 atb, Perisic, no 11.5 FWD, premium GK, rotating GKs, onemium, threemium, 0 ITB, Matip?
I'm thrilled to congratulate Jonas (
@ASC_FPL
) on his historic Fantasy Premier League win. Jonas, your victory as the first champion from the analytics community marks a significant milestone in our shared passion for data-driven success in sports. You've made us all proud!
#FPL
A first look at raw no-transfer optimal team for GW17-24 using Review projections.
As usual, premium GK. Nice DGW coverages. Will start tinkering from here.
#FPL
While EV Junkies chat is going on, I generated optimal FH32 teams using raw data from various sources.
The first one is optimal of each data source by itself. The last two columns are "Review/Mikkel Mix" (50/50) and ensemble (all equal weights). Variance is based on own %.
#FPL
Lots of topics are being discussed from BB1 to premium GKs. Some tinkering is OK, but few people seem to insist they are right. Even if you are, why are you trying to make people join you? Isn't it better to stay quiet about it?
#FPL
Another mid-GW brag for you: Analytics Discord league (Bullet Wisdom) has 39 players in top 10K, with 12 of them are within 1K.
Total number of teams in this league is only 182.
Analytics season. 😎