1 of 19

Aramis Farias / April 28

Empirical Advantages on Chess Openings

The importance of a good start

Proprietary + Confidential

2 of 19

What is a chess opening?

Proprietary + Confidential

3 of 19

Among hundreds, which openings to focus the learning?

Problem:

Provide a data-driven approach to scope for openings that empirically offer advantages to the player.

Solution:

Proprietary + Confidential

4 of 19

Which openings to learn in chess?

Chess-360 currently offers a conventional approach to openings:

  • Based on frequency of openings.
  • Doesn’t prepare for unusual openings to avoid or exploit.
  • Too conventional, so adversaries are probably also aware.

Proprietary + Confidential

5 of 19

Which openings to learn in chess?

Here we present a innovative data-driven approach:

  • Based on Win-loss ratios of high level players.
  • Controlled for players skill and game format.
  • Competitive Edge through unconventional perspective.

Proprietary + Confidential

6 of 19

Methodology

Filters

Categories

Separation by winning condition, a.k.a Termination

Separation by format

•Small difference in skill (Elo dif<50)

•Only skilled players (Elo>2000)

6 million games dataset

•At least 100 games per opening

Proprietary + Confidential

7 of 19

Game outcomes according to the Opening

•••

All visualizations available here

•••

Proprietary + Confidential

Proprietary + Confidential

8 of 19

Best Winning rates for White per Opening and Format

All visualizations available here

Blitz (10 min > time >= 3 min):

  • C22 - 57,1% - Center Game
  • C61 - 56,5% - Ruy Lopez, Bird's Defense
  • E32 - 56,2% - Nimzo-Indian, Classical

Bullet (time < 3 min):

  • A30 - 59,9% - English, Symmetrical
  • C27 - 56,2% - Vienna Game
  • D13 - 56,2% - Queen's Gambit Declined Slav, Exchange Variation

Rapid (30 min >= time >= 10 min):

  • C30 - 61,1% - King's Gambit Declined
  • C53 - 57,4% - Giuoco Piano
  • C21 - 57,0% - Center Game

Proprietary + Confidential

Proprietary + Confidential

9 of 19

Best Winning rates for Black per Opening and Format

All visualizations available here

Blitz (10 min > time >= 3 min):

  • C51 - 55,9% - Evans Gambit
  • A60 - 55,1% - Benoni Defense
  • E65 - 54,7% - King's Indian, Fianchetto, Yugoslav

Bullet (time < 3 min):

  • C46 - 56,5% - Three Knights
  • A53 - 55,8% - Old Indian
  • A51 - 55,4% - Budapest Gambit

Rapid (30 min >= time >= 10 min):

  • B03 - 57,3% - Alekhine's Defense
  • C20 - 54,9% - King's Pawn Game
  • C50 - 51,8% - Giuoco Piano

Proprietary + Confidential

Proprietary + Confidential

10 of 19

Predicted Benefit of the openings found

White:

  • It can be expect a winning chance of around 56,5%(18% increase) for white. This advantage can be translated in a 63 points advantage in ELO rating, or 2 pawns advantage in game material.

Black:

  • It can be expect a winning chance of around 55,2%(15% increase) for black. This advantage can be translated in a 54 points advantage in ELO rating, or 2 pawns advantage in game material.

Note: The regular probability of victory for even skilled players is 47,7%

Proprietary + Confidential

Proprietary + Confidential

11 of 19

Best Winning rates per Termination

For White:

  • D13 - 62,5% - Queen's Gambit Declined Slav, Exchange Variation
  • D32 - 59,4% - Queen's Gambit Declined, Tarrasch
  • B80 - 56,9% - Sicilian, Scheveningen

For Black:

  • A42 - 57,3% - Modern Defense, Averbakh System
  • B09 - 56,3% - Pirc, Austrian Attack
  • B43 - 55,1% - Sicilian, Kan, 5.Nc3

Proprietary + Confidential

Proprietary + Confidential

12 of 19

Predicted Benefit of seeking time forfeit

White:

  • It can be expect a winning chance of around 59,6%(25% increase) for white. This advantage can be translated in a 86 points advantage in ELO rating, or 3 pawns advantage in game material.

Black:

  • It can be expect a winning chance of around 56,2%(17% increase) for black. This advantage can be translated in a 61 points advantage in ELO rating, or 2 pawns advantage in game material.

Note: this would be mainly useful in formats with less time, namely Blitz and Bullet.

Proprietary + Confidential

Proprietary + Confidential

13 of 19

Comparison with the most frequent openings

Top 10 Most Frequent Openings

  • A00 6,7% - Uncommon Opening
  • A40 4,6% - Queen's Pawn Game
  • C00 4,6% - French Defense
  • B01 4,5% - Scandinavian
  • D00 3,6% - Queen's Pawn Game
  • B00 3,3% - Uncommon King's Pawn Opening
  • C41 3,1% - Philidor Defense
  • C20 2,8% - King's Pawn Game
  • B20 2,3% - Sicilian
  • D02 2,0% - Queen's Pawn Game

9

23

1

Most Frequent Openings

New openings recommended

Only 1 intersection

Proprietary + Confidential

Proprietary + Confidential

14 of 19

Key Findings

New approach

Synergy

Time strategy

The study of the openings found here present opportunities to exploit or avoid, which can increase the winning chance by even 25% against an even skilled player.

The majority of the openings found isn’t portrayed in the conventional study of openings. We suggest a complementary study of both approaches.

Surprisingly, there was consistency in some openings forcing the time forfeit, offering better odds than the usual in Blitz and Bullet games.

This suggest a new strategy to be implemented

Proprietary + Confidential

15 of 19

Thank You

Proprietary + Confidential

16 of 19

Appendix

Proprietary + Confidential

17 of 19

Reference

Proprietary + Confidential

18 of 19

Data Sources

  1. Chess Game Dataset (Lichess) - https://www.kaggle.com/datasnaek/chess

  • Chess_dataset (100,000 games) Lichess - https://www.kaggle.com/sahit2509/chess-dataset-100000-games-lichess

  • Chess Games - https://www.kaggle.com/arevel/chess-games

Proprietary + Confidential

19 of 19

Supplementary tools used

  • Elo Win Probability Calculator - https://wismuth.com/elo/calculator.html

  • Chess opening names - https://www.chessgames.com/perl/chessopening?eco=D02

Proprietary + Confidential