Datasets:
β Chess Evaluation-Based Feature Dataset
Dataset Description
This dataset contains over 5 million chess games sourced from Lichess.org, containing Stockfish Engine Evaluations and clock timings and moves and game metadata.
The dataset is composed of two files:
lichess_games_moves: Contains traditional PGN-style metadata and move sequences, stored in Parquet format for fast querying.lichess_moves_sample.parquet: Contains per-player, per-game performance metrics derived from evaluations and timing data, including centipawn loss, move accuracy, and time usage.
Each game is uniquely identified by a game_id (UUID), which allows for seamless joins between the two datasets.
This structure enables researchers to analyze player behavior at both the game level and the player level, supporting tasks such as cheat detection, player profiling, skill progression analysis, and AI performance benchmarking. The dataset spans a wide range of game types, rating tiers, and event formats, offering a granular view into decision-making, precision, and time management across millions of chess games.
π Dataset Summary
| Attribute | Description |
|---|---|
| Total Games | 5,000,000 |
| ELO Range | 400 to 3945 |
| Time Controls | Bullet (21%), Blitz (47%), Rapid (29%), Classical (2%), Ultrabullet (0.7%) |
| Unique Openings | 142 openings, 100 variations |
| Event Types | Game (88%), Tournament (11%), Swiss Tournament (1%) |
π Features
| Field | Description | Calculation |
|---|---|---|
| total_cpl | Total centipawn loss | sum(abs(eval_after - eval_before)) |
| acpl | Average centipawn loss | total_cpl / move_count |
| mcpl | Median centipawn loss | median(abs(eval_after - eval_before)) |
| std_cpl | Std deviation of CPL | std(abs(eval_after - eval_before)) |
| 25%_cpl | 25th percentile CPL | percentile(abs(eval_after - eval_before), 25) |
| 75%_cpl | 75th percentile CPL | percentile(abs(eval_after - eval_before), 75) |
| min_cpl | Minimum CPL | min(abs(eval_after - eval_before)) |
| max_cpl | Maximum CPL | max(abs(eval_after - eval_before)) |
| move_cnt | Count of evaluated moves | Count of moves with evaluations |
| inaccuracy_cnt | Moves with win prob drop 5β10% | 5 < delta_wp < 10 |
| mistake_cnt | Moves with win prob drop 10β15% | 10 < delta_wp < 15 |
| blunder_cnt | Moves with win prob drop β₯15% | delta_wp β₯ 15 |
| best_cnt | Moves with win prob drop β€5% | delta_wp β€ 5 |
| perfect_streak | Longest best-move streak | max(consecutive_moves where delta_wp β€ 5) |
| mean_acc | Mean move accuracy | mean(accuracy_scores) |
| med_acc | Median accuracy | median(accuracy_scores) |
| std_acc | Std deviation of accuracy | std(accuracy_scores) |
| 25%_acc | 25th percentile accuracy | percentile(accuracy_scores, 25) |
| 75%_acc | 75th percentile accuracy | percentile(accuracy_scores, 75) |
| min_acc | Minimum accuracy | min(accuracy_scores) |
| max_acc | Maximum accuracy | max(accuracy_scores) |
| mean_win% | Mean win probability | mean(win_probs) |
| med_win% | Median win probability | median(win_probs) |
| std_win% | Std deviation win prob | std(win_probs) |
| 25%_win% | 25th percentile win prob | percentile(win_probs, 25) |
| 75%_win% | 75th percentile win prob | percentile(win_probs, 75) |
| min_win% | Minimum win probability | min(win_probs) |
| max_win% | Maximum win probability | max(win_probs) |
| avg_tpm | Avg time per move (sec) | mean(move_times) |
| med_tpm | Median time per move | median(move_times) |
| stdev_tpm | Std deviation of time | std(move_times) |
| 25%_tpm | 25th percentile time | percentile(move_times, 25) |
| 75%_tpm | 75th percentile time | percentile(move_times, 75) |
| min_tpm | Minimum time per move | min(move_times) |
| max_tpm | Maximum time per move | max(move_times) |
| instant_move_cnt | Moves made in β€1s | count(time β€ 1) |
| game_id | Unique game identifier | UUID format |
| player | Player color | white or black |
| username | Player username | From Lichess |
| elo | Player rating | ELO score |
| rating_diff | Rating change | Post-game change |
| title | Chess title | e.g., GM, IM, CM |
| Site | Game URL | Lichess link |
| Date | Local game date | YYYY.MM.DD |
| Round | Game round | Often - |
| Result | Game result | 1-0, 0-1, or 1/2-1/2 |
| UTCDate | UTC date | Game timestamp |
| UTCTime | UTC time | Game start time |
| ECO | ECO code | e.g., A00, B01 |
| Opening | Opening name | Named variation |
| TimeControl | Raw control format | e.g., 600+0 |
| Termination | End reason | e.g., Normal, Timeout |
| timecontrol_type | Time class | Bullet, Blitz, etc. |
| event_type | Event format | Game, Tournament, Swiss |
| event_url | Tournament link | If applicable |
Win Probability: WP(cp) = 50 + 50 Γ ( 2 / (1 + e^(β0.00368208 Γ cp)) β 1 )
Accuracy: ΞWP = WP_before β WP_after Accuracy(ΞWP) = clamp( 103.1668 Γ e^(β0.04354 Γ ΞWP) β 3.1669 , 0, 100 )
π Sample Observation from lichess_metrics_sample.parquet
{
"total_cpl": 1436.0,
"acpl": 27.615384615384617,
"mcpl": 10.0,
"std_cpl": 75.82777901934035,
"25%_cpl": 4.0,
"75%_cpl": 16.25,
"min_cpl": 0.0,
"max_cpl": 474.0,
"move_cnt": 52,
"inaccuracy_cnt": 0,
"mistake_cnt": 0,
"blunder_cnt": 2,
"best_cnt": 50,
"perfect_streak": 37,
"mean_acc": 95.04878233982011,
"med_acc": 99.06330947144633,
"std_acc": 14.906050942960489,
"25%_acc": 97.21119376075846,
"75%_acc": 100.0,
"min_acc": 16.396635484120793,
"max_acc": 100.0,
"mean_win%": 75.49193923600627,
"med_win%": 85.45913904456496,
"std_win%": 18.60559724644749,
"25%_win%": 52.46040066101623,
"75%_win%": 87.16445709539632,
"min_win%": 35.17797456776559,
"max_win%": 97.54474363414323,
"avg_tpm": 13.257142857142858,
"med_tpm": 12.0,
"stdev_tpm": 7.904222586306483,
"25%_tpm": 7.0,
"75%_tpm": 20.0,
"min_tpm": 2.0,
"max_tpm": 29.0,
"instant_move_cnt": 0,
"game_id": "3c2687ff-b368-4ab0-8018-05346ccd16e9",
"player": "white",
"username": "TIBURONCHILENO",
"elo": "2366",
"rating_diff": "+1",
"title": "FM",
"Site": "https://lichess.org/v0oGqi2g",
"Date": "2025.02.20",
"Round": "-",
"Result": "1-0",
"UTCDate": "2025.02.20",
"UTCTime": "00:48:50",
"ECO": "A45",
"Opening": "Indian Defense",
"TimeControl": "180+2",
"Termination": "Normal",
"timecontrol_type": "Blitz",
"event_type": "tournament",
"event_url": "https://lichess.org/tournament/oeFqsJi9",
"moves":
}
π Sample Observation from lichess_moves_sample.parquet
{
"game_id": "3c2687ff-b368-4ab0-8018-05346ccd16e9",
"Site": "https://lichess.org/v0oGqi2g",
"Date": "2025.02.20",
"Round": "-",
"White": "TIBURONCHILENO",
"Black": "Gran_Maestro2756",
"Result": "1-0",
"UTCDate": "2025.02.20",
"UTCTime": "00:48:50",
"WhiteElo": "2366",
"BlackElo": "1940",
"WhiteRatingDiff": "+1",
"BlackRatingDiff": "-1",
"ECO": "A45",
"Opening": "Indian Defense",
"TimeControl": "180+2",
"Termination": "Normal",
"BlackTitle": None,
"WhiteTitle": "FM",
"event_url": "https://lichess.org/tournament/oeFqsJi9",
"timecontrol_type": "Blitz",
"event_type": "tournament",
"has_eval": "1",
"moves": "1. d4 { [%eval 0.17] [%clk 0:03:00] } 1... Nf6 { [%eval 0.19] [%clk 0:03:00] } 2. Bf4 { [%eval 0.05] [%clk
0:03:01] } 2... e6 { [%eval 0.11] [%clk 0:03:01] } 3. e3 { [%eval 0.0] [%clk 0:03:03] } 3... c5 { [%eval 0.0] [%clk 0:03:02] }
4. Nd2 { [%eval 0.1] [%clk 0:03:04] } 4... cxd4 { [%eval 0.37] [%clk 0:03:03] } 5. exd4 { [%eval 0.15] [%clk 0:03:04] } 5...
b6 { [%eval 0.19] [%clk 0:03:04] } 6. c3 { [%eval 0.16] [%clk 0:03:06] } 6... Bb7 { [%eval 0.2] [%clk 0:03:05] } 7. Ngf3 {
[%eval 0.15] [%clk 0:03:07] } 7... d6 { [%eval 0.27] [%clk 0:03:06] } 8. Bd3 { [%eval 0.25] [%clk 0:03:08] } 8... Be7 { [%eval
0.29] [%clk 0:03:07] } 9. Qe2 { [%eval 0.26] [%clk 0:03:10] } 9... O-O { [%eval 0.24] [%clk 0:03:07] } 10. h3 { [%eval 0.28]
[%clk 0:03:11] } 10... Nbd7 { [%eval 0.27] [%clk 0:03:08] } 11. O-O { [%eval 0.27] [%clk 0:03:13] } 11... Re8 { [%eval 0.25]
[%clk 0:03:09] } 12. Rfe1 { [%eval 0.28] [%clk 0:03:14] } 12... Qc7 { [%eval 0.28] [%clk 0:03:10] } 13. Ng5 { [%eval 0.14]
[%clk 0:03:15] } 13... h6? { [%eval 1.37] [%clk 0:03:10] } 14. Nxe6?? { [%eval -1.66] [%clk 0:03:13] } 14... fxe6 { [%eval
-1.39] [%clk 0:03:04] } 15. Qxe6+ { [%eval -1.38] [%clk 0:03:15] } 15... Kf8?? { [%eval 3.98] [%clk 0:03:05] } 16. Bg6?? {
[%eval -0.76] [%clk 0:03:15] } 16... Bd5 { [%eval -0.75] [%clk 0:02:58] } 17. Qf5 { [%eval -0.51] [%clk 0:03:16] } 17... Red8?
{ [%eval 2.48] [%clk 0:02:57] } 18. c4 { [%eval 2.62] [%clk 0:03:09] } 18... Bf7? { [%eval 4.93] [%clk 0:02:57] } 19. Bxf7 {
[%eval 4.89] [%clk 0:02:50] } 19... Kxf7 { [%eval 4.91] [%clk 0:02:58] } 20. Qe6+ { [%eval 4.88] [%clk 0:02:52] } 20... Kg6 {
[%eval 4.86] [%clk 0:02:58] } 21. Qxe7 { [%eval 4.64] [%clk 0:02:51] } 21... Re8 { [%eval 4.87] [%clk 0:02:56] } 22. Qxd6 {
[%eval 4.97] [%clk 0:02:52] } 22... Qxd6 { [%eval 4.86] [%clk 0:02:57] } 23. Bxd6 { [%eval 4.92] [%clk 0:02:45] } 23... Rad8 {
[%eval 5.05] [%clk 0:02:56] } 24. Bc7 { [%eval 4.95] [%clk 0:02:43] } 24... Rc8 { [%eval 4.97] [%clk 0:02:56] } 25. Bg3 {
[%eval 4.8] [%clk 0:02:45] } 25... Nh5 { [%eval 4.99] [%clk 0:02:55] } 26. Bd6 { [%eval 5.09] [%clk 0:02:39] } 26... Nhf6 {
[%eval 4.95] [%clk 0:02:53] } 27. f3 { [%eval 4.82] [%clk 0:02:40] } 27... Rcd8 { [%eval 5.1] [%clk 0:02:46] } 28. Kf2 {
[%eval 4.96] [%clk 0:02:38] } 28... Nf8 { [%eval 5.29] [%clk 0:02:41] } 29. Rxe8 { [%eval 5.19] [%clk 0:02:37] } 29... Rxe8 {
[%eval 5.25] [%clk 0:02:41] } 30. Bxf8 { [%eval 5.0] [%clk 0:02:33] } 30... Rxf8 { [%eval 4.9] [%clk 0:02:42] } 31. Re1 {
[%eval 4.86] [%clk 0:02:34] } 31... Rd8 { [%eval 5.0] [%clk 0:02:43] } 32. d5 { [%eval 5.03] [%clk 0:02:36] } 32... Nd7 {
[%eval 5.37] [%clk 0:02:43] } 33. Ne4 { [%eval 4.42] [%clk 0:02:35] } 33... Ne5 { [%eval 4.39] [%clk 0:02:43] } 34. Rd1 {
[%eval 3.9] [%clk 0:02:14] } 34... Nxc4 { [%eval 3.75] [%clk 0:02:43] } 35. b3 { [%eval 3.66] [%clk 0:02:15] } 35... Nd6 {
[%eval 4.41] [%clk 0:02:41] } 36. Nxd6 { [%eval 4.36] [%clk 0:02:16] } 36... Rxd6 { [%eval 4.13] [%clk 0:02:42] } 37. Ke3 {
[%eval 4.35] [%clk 0:02:17] } 37... Kf5 { [%eval 4.57] [%clk 0:02:42] } 38. Kd4 { [%eval 4.42] [%clk 0:02:18] } 38... a6 {
[%eval 4.89] [%clk 0:02:41] } 39. Re1 { [%eval 4.77] [%clk 0:02:19] } 39... g6 { [%eval 5.07] [%clk 0:02:41] } 40. g3 { [%eval
4.92] [%clk 0:02:19] } 40... h5 { [%eval 5.35] [%clk 0:02:39] } 41. h4 { [%eval 5.32] [%clk 0:02:19] } 41... Kf6 { [%eval 5.3]
[%clk 0:02:38] } 42. Re8 { [%eval 5.36] [%clk 0:02:18] } 42... Kf7 { [%eval 5.08] [%clk 0:02:39] } 43. Rc8 { [%eval 5.24]
[%clk 0:02:19] } 43... Rf6 { [%eval 5.8] [%clk 0:02:35] } 44. f4 { [%eval 5.76] [%clk 0:02:19] } 44... Ke7 { [%eval 5.63]
[%clk 0:02:30] } 45. b4 { [%eval 5.7] [%clk 0:02:10] } 45... a5 { [%eval 5.93] [%clk 0:02:29] } 46. b5 { [%eval 5.82] [%clk
0:02:11] } 46... Kd7 { [%eval 5.89] [%clk 0:02:27] } 47. Rc6 { [%eval 5. 85] [%clk 0:02:11] } 47... Rd6?? { [%eval 9.74] [%clk
0:02:20] } 48. Ke5 { [%eval 9.51] [%clk 0:02:11] } 48... Rxc6 { [%eval 9. 77] [%clk 0:02:18] } 49. bxc6+ { [%eval 9.58] [%clk
0:02:13] } 49... Kc7 { [%eval 10.01] [%clk 0:02:18] } 50. Ke6 { [%eval 9. 37] [%clk 0:02:15] } 50... Kd8 { [%eval 10.67] [%clk
0:02:19] } 51. d6 { [%eval 10.36] [%clk 0:02:14] } 51... b5 { [%eval 10. 49] [%clk 0:02:19] } 52. d7 { [%eval 10.26] [%clk
0:02:16] } 52... b4?! { [%eval #2] [%clk 0:02:19] } 53. Kd6 { [%eval #1] [%clk 0:02:17] } 53... b3 { [%eval #1] [%clk 0:02:20]
} 54. c7# { [%clk 0:02:17] } 1-0"
}
Dataset Sources
Lichess: [https://database.lichess.org/]
Dataset Card Contact
If you are interested in accessing the full dataset, reach out to [email protected]. Proceeds from the sales of the full dataset will be directed to the The Gift of Chess Inc a 501(c)(3) nonprofit organization transforming lives through the universal language of chess.
We are currently expanding our augmented dataset offerings and are working closely with data consumers to match them directly to high-quality real-world, augmented, and synthetic data sources.
Whether you are pre or post-training models, fine tuning, conducting evals or just looking for novel datasets to assist in your AI development, Portex would like to partner with you.
- Downloads last month
- 9