• 

  A generated batch file of the team XI.

• 

  A generated png image of the team XI.

TFF Fantasy League Optimiser

Tagline: Optimal TFF Fantezi Lig squads from form, fixtures, and integer programming — not vibes.

Built with: Python · PuLP · Pandas · Sofascore · FotMob · TFF Fantezi Lig API · Pillow

Inspiration

Fantasy football in Turkey’s Süper Lig is intense: 100M budget, strict squad rules, weekly fixtures, and scoring that rewards minutes, clean sheets, goals/assists, and bonuses. Most tools stop at price lists or gut picks. We wanted something that actually answers: given this week’s opponents and real form, what’s the best legal 15-man squad? Especially early in the season — when one good match can look like a season — we needed Bayesian-style caution, not hype.

What it does

TFF Fantasy League Optimiser builds a weekly recommended squad for the official TFF Fantezi Lig:

• Pulls player prices and availability from TFF
• Estimates expected points from Sofascore + FotMob (form, season stats, xG/xA, minutes)
• Respects official scoring (60+ minutes, CS, cards, bonuses, etc.)
• Solves an integer program for the best formation, starting XI, bench, and captain under 100M / 2-5-5-3 / max 3 per club
• Advises whether to play a manager card or save it
• Exports a clean weekly PNG report (pitch, fixtures, card call)

One command (calistir.bat or python -m src.main) → ready-to-use lineup.

How we built it

Python end-to-end pipeline:

1. Ingest — TFF client for prices/availability; Sofascore & FotMob for performance (cached)
2. Score — Per-player expected points with opponent strength, home/away, early-season blending of prior season when sample size is tiny
3. Calibrate — External-league → Süper Lig translation from historical transfers (league_translation.json), not a flat “foreign = weaker” guess
4. Optimize — PuLP integer program over formations; weighted bench for auto-subs
5. Advise — Manager-card opportunity ranking vs. saving the card
6. Report — Pillow-generated weekly pitch graphic

Challenges we ran into

• Early-season noise: One clean sheet can look like 100% CS probability; unplayed clubs look like 0%. We had to shrink small samples and blend with last season.
• Same-club goalkeepers: Without start probability, a backup with one lucky CS could beat the real #1.
• New transfers / foreign leagues: Mapping Ligue 1 or Championship rates into Super Lig TFF points needed real transfer history, not hand-waved multipliers.
• Name matching & availability: TFF vs Sofascore vs FotMob naming, plus injured/suspended exclusions.
• Card advice without inventory: We rank opportunity value (use vs save), not account stock.

Accomplishments that we're proud of

• A full legal fantasy squad in one run — not just a ranked list
• Transparent reasons per player (why this XP, start %, CS blend)
• Robust early-season / GK / form fallback logic that doesn’t overfit week 1
• League translation calibrated on real Super Lig arrivals
• A weekly PNG report you can actually share or screenshot into decisions

What we learned

Expected points ≠ last match points. Sample size, role (starter vs backup), fixture difficulty, and prior-season priors matter more than a viral one-off performance. Also: fantasy scoring is a product of team defense and individual minutes — modeling only goals misses half the game for GKs and defenders.

What's next for TFF Fantasy League Optimiser

• Live / confirmed XI signals when available
• Multi-week transfer planning (bank, hits, chip windows)
• Stronger integration of official TFF points for mid-season calibration
• Optional web UI on top of the same optimizer
• More automated regression checks on “week 1 traps” (hot GKs, zero-sample clubs)

Built With

• cursor
• fantasyleague
• football
• fotmob
• pandas
• sofascore
• tff
• ts