Skip to content

CHANGELOG.md

Latest commit

 

History

History
96 lines (84 loc) · 5.25 KB

CHANGELOG.md

File metadata and controls

96 lines (84 loc) · 5.25 KB
Raw

Changelog

Notable changes to this package. Versions follow semantic versioning.

0.3.0

find_arbitrage and find_value_bets now work on markets carrying a line.

Breaking: find_value_bets takes devig where it took sharp, and window_hours is gone from group_games, find_arbitrage and find_value_bets.

  • Totals, handicaps, team totals and player props are supported. Outcomes are grouped into individual bets before being paired, so Over 8.5 is never matched against Under 9.5, one player's prop never against another's, and one team's total never against the other team's
  • Handicaps pair a team against its opponent on the mirrored line. Books disagree on who is giving start, so -1.5 and +1.5 on the same number are two separate bets and backing the same side at two books is not an arb
  • Which kind of market a response holds is read from the outcomes themselves, not from a list of market keys, so a market the API adds later works without an SDK release
  • ArbLeg and ValueBet carry point on any market with a line, player_name on player props and team_name on team totals. Head to head results are unchanged, since none of the three applies
  • One result per line, so a totals market with ten lines can return ten
  • find_arbitrage no longer discards pairs whose implied probabilities sum to 1 or more, leaving min_profit as the only filter, so a negative value now shows near misses. At the default of 0.0 the only new result is a pair summing to exactly 1, reported as a 0% arb
  • find_value_bets takes devig in place of sharp, and the fair price can now come from more than one book. devig="Pinnacle" is the old behaviour, devig="all" averages every book in the response, and devig={"Pinnacle": 0.8, "TAB": 0.2} averages the books you name at your own weights. Each book is de-vigged on its own before the fair odds are averaged, weighted by 1 / (1 + margin) ** 2 unless you supply weights
  • min_books on find_value_bets, four by default, is how many books have to price a bet before a consensus is trusted. Naming books lowers it to the number named. It does not apply when de-vigging a single book
  • Under a consensus no book is excluded from the results, so both sides of a game can show value at once. A single book is still excluded, since it cannot beat its own price
  • find_value_bets skips a book whose own two prices sum to under 1. That is the book arbing itself, which is a bad feed rather than a thin margin, and normalising it would have invented edge at every other book
  • Under a consensus, a fair price outside 1.1 to 50, or an edge over 50%, is discarded as bad data. One stale price drags an average in a way it cannot drag a single book's de-vig, so these bounds apply to consensus only
  • A handicap outcome naming neither team is ignored. Books disagree on team names, and one quoting the Oakland Athletics as "Athletics" would otherwise read as a third side and make every line of that game unpairable
  • A handicap of zero is ignored. It pushes when the teams finish level, so both sides come back rather than one winning, and the pair is not a locked profit
  • Only three-way markets now raise ValidationError, recognised by a draw or tie among the outcomes rather than by counting them. A bet left with more than two sides for any other reason is skipped, so one book's naming cannot take down a whole market
  • The window for matching a game across bookmakers now comes from the sport, and window_hours is gone from group_games, find_arbitrage and find_value_bets. MLB gets 1.8 hours and everything else 6, in MATCH_WINDOW_HOURS. The old flat 4 hours merged a split doubleheader starting three hours after the first game, which compared prices across two different games. match_window reports what a response will get

0.2.0

Four methods for things the API can now answer directly. None of them cost credits, and all need the API at v1.1.0 or later.

  • list_sports and get_sport, which return the sports the API accepts and, on request, each one's market keys. The market keys are what get_odds takes as market_types, so they no longer have to be looked up by hand
  • list_results_sports for the sports carrying live scores and player stats
  • get_usage for the credit balance the server holds. Unlike credits_used it survives restarts and counts spend from anywhere, not just this client
  • SportInfo, SportMarkets and Usage response types

The sport tuples in constants stay as an offline snapshot, and BOOKMAKERS remains the only list of bookmaker feeds, since the API has no endpoint for them.

0.1.0

First release.

  • RapidOddsAPI and AsyncRapidOddsAPI clients over REST
  • get_odds and get_results, with full type hints on every response
  • stream_odds and stream_results over WebSocket, with automatic reconnect and re-subscribe
  • find_arbitrage and find_value_bets for two-way head to head markets
  • group_games to match a game across bookmakers on teams plus a time window
  • parse_time for the API's naive UTC timestamps
  • Typed exceptions, including separate classes for rate limiting and running out of credits
  • Automatic retry with backoff on 5xx, rate limits and network failures
  • credits_used tally per client