What poker data can I download?
7 datasets, each as a JSON file and a CSV file. They are the same numbers the calculators and the data pages on this site use, computed by our own code from the 52-card deck and a solver, with no third-party data. Open them in a spreadsheet, a notebook or your own tool.
| Dataset | What it holds | Where it appears |
|---|---|---|
| Preflop hand-vs-hand equity | 28,561 matchups, exact to 5 decimals | Hand matchups |
| Starting hands ranked by equity | 169 hands with rank, combos and percentile | Hand combos |
| Heads-up push or fold charts | 6,591 rows: 39 stacks by 169 hands | Push or fold charts |
| ICM examples for 3 to 6 players | One row per seat in each of 38 examples | ICM examples |
| Losing streak odds | 27 rows by win chance, trials and run length | Mental game guides |
| Pot odds by bet size | 12 bet sizes from a quarter pot to 4 times the pot | Pot odds by bet size |
| Outs and the chance to hit | 40 rows: 20 outs on two streets | Outs by street |
What is the license?
Every file is published under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). You may copy, share and build on the data, including in paid work, as long as you give credit with the line "George Li, georgelipoker.com". The JSON files carry the license and the credit inside them, so a file keeps both when it is copied away from this page.
How do I cite the data?
Use the credit line and name the file, for example: "Preflop equity, preflop-equity.csv: George Li, georgelipoker.com/data, CC BY 4.0." For a paper or a chart, add the date you downloaded the file, because the data is regenerated when the method improves.
What does each dataset contain?
| Dataset | Columns (CSV) |
|---|---|
| preflop-equity | hero, villain, win, tie, lose, equity |
| hand-ranking | rank, hand, combos, equity_vs_random, percentile |
| push-fold-heads-up | stack_bb, hand, small_blind_shove_frequency, big_blind_call_frequency, small_blind_shove_ev_bb, big_blind_call_ev_bb, small_blind_equity_when_called, big_blind_equity_vs_shoves |
| icm-examples | players, structure, pattern, payouts_percent, seat, chips, chip_share_percent, icm_equity_percent, icm_minus_chips_points, lowest_bubble_factor, highest_bubble_factor |
| losing-streak-odds | win_chance, trials, run_length, chance |
| pot-odds | bet_slug, bet_as_fraction_of_pot, equity_needed, price |
| outs | street, outs, unseen_cards, hit_by_river, hit_next_card, rule, rule_percent, rule_error_points |
Preflop hand-vs-hand equity: Win, tie and lose chances for every pair of the 169 starting-hand classes, each averaged over every pair of specific hands that share no card and every five-card board.
Starting hands ranked by equity: The 169 starting-hand classes from strongest to weakest by equity against a random hand, with combo counts and the share of all 1,326 hands held by each class and every stronger one.
Heads-up push or fold charts: The small blind shove frequency and the big blind call frequency for every starting hand at every stack from 1 to 20 big blinds in half steps, with the EV of acting and the equity behind it. Chips only, no ante, solved to an exploitability of at most 0.001 big blind.
ICM examples for 3 to 6 players: Malmuth-Harville ICM equity for every seat in 38 stack patterns, with two payout structures, plus the lowest and highest bubble factor each seat faces.
Losing streak odds: The exact chance of at least one run of losses in a row of a given length within a number of independent trials, for win chances of 40, 50 and 60 percent.
Pot odds by bet size: The equity a call needs for twelve bet sizes, f / (1 + 2f) for a bet of f times the pot, and the price the pot offers.
Outs and the chance to hit: The exact chance to hit 1 to 20 outs on the flop by the river and on the turn, the chance the next card is an out, and how far the rule of 4 and 2 is from the exact figure.
Where can I see the data as pages?
Each dataset feeds a family of pages that read it row by row and explain what the numbers mean. Open the family for the readable version of a file.
How was the data computed?
Equity counts every five-card board for every pair of specific hands that share no card. The push or fold charts come from a heads-up solver that plays itself to an exploitability of at most 0.001 big blind. ICM uses the Malmuth-Harville model, losing streak odds are exact by dynamic programming, and pot odds and outs are closed formulas, so the same question always returns the same number.
Source: GeorgeLiPoker's own computation from the 52-card deck.
Questions
- Can I use this poker data in a paid product?
- Yes. CC BY 4.0 allows commercial use. Give credit with the line "George Li, georgelipoker.com" and say if you changed the data.
- Is there an API for the data?
- No. The datasets are static files at fixed addresses, so you can download them once or fetch them from a script. Each file is also readable page by page in the matching data family.
- Which file should I start with for preflop equity?
- Start with the CSV of preflop equity: one row per matchup of two of the 169 starting-hand classes, with the win, tie, lose and equity columns. The JSON version holds the same matrix in a compact grid.