Initial commit

This commit is contained in:
2026-09-17 21:45:57 +02:00
commit 9160e861e0
302 changed files with 31275 additions and 0 deletions

View File

@@ -0,0 +1,139 @@
-- =====================================================================
-- OpenAI match predictions + comparison views
-- Run against the "bookie" schema (Postgres).
-- Safe to run multiple times (IF NOT EXISTS / CREATE OR REPLACE).
-- =====================================================================
-- ---------------------------------------------------------------------
-- 1) Table: one row PER TEAM PER EXECUTION of the OpenAI predictor.
-- History is kept (no unique constraint) so you can inspect how
-- predictions change across runs; the views below always pick the
-- most recent row per (match, team).
-- ---------------------------------------------------------------------
CREATE TABLE IF NOT EXISTS bookie.match_prediction_openai (
prediction_id SERIAL PRIMARY KEY,
match_id INT NOT NULL REFERENCES bookie.match(match_id),
team_id INT NOT NULL REFERENCES bookie.team(team_id),
predicted_goals NUMERIC(6,2),
goals_low NUMERIC(6,2),
goals_high NUMERIC(6,2),
predicted_shots_on_target NUMERIC(6,2),
shots_on_target_low NUMERIC(6,2),
shots_on_target_high NUMERIC(6,2),
predicted_corners NUMERIC(6,2),
corners_low NUMERIC(6,2),
corners_high NUMERIC(6,2),
predicted_fouls NUMERIC(6,2),
fouls_low NUMERIC(6,2),
fouls_high NUMERIC(6,2),
predicted_yellow_cards NUMERIC(6,2),
yellow_cards_low NUMERIC(6,2),
yellow_cards_high NUMERIC(6,2),
predicted_red_cards NUMERIC(6,2),
red_cards_low NUMERIC(6,2),
red_cards_high NUMERIC(6,2),
confidence TEXT,
reasoning TEXT,
model TEXT,
predicted_at TIMESTAMP NOT NULL DEFAULT NOW()
);
CREATE INDEX IF NOT EXISTS idx_match_prediction_openai_match_team
ON bookie.match_prediction_openai (match_id, team_id, predicted_at DESC);
-- ---------------------------------------------------------------------
-- 2) Latest OpenAI prediction per (match, team).
-- ---------------------------------------------------------------------
CREATE OR REPLACE VIEW bookie.v_match_prediction_openai_latest AS
SELECT DISTINCT ON (match_id, team_id) *
FROM bookie.match_prediction_openai
ORDER BY match_id, team_id, predicted_at DESC;
-- ---------------------------------------------------------------------
-- 3) Side-by-side comparison: LLM (bookie.match_prediction) vs OpenAI
-- vs ACTUAL, one row per (match, team). Actual goals come from the
-- final score; the other actuals come from bookie.match_team_stats.
-- ---------------------------------------------------------------------
CREATE OR REPLACE VIEW bookie.v_prediction_comparison AS
SELECT
m.match_id,
t.team_id,
t.name AS team_name,
m.kickoff_at,
m.status,
(m.home_team_id = t.team_id) AS is_home,
opp.name AS opponent_name,
-- goals
lp.predicted_goals AS llm_goals,
op.predicted_goals AS openai_goals,
CASE WHEN m.home_team_id = t.team_id THEN m.home_score_ft ELSE m.away_score_ft END AS actual_goals,
-- shots on target
lp.predicted_shots_on_target AS llm_shots_on_target,
op.predicted_shots_on_target AS openai_shots_on_target,
mts.shots_on_target AS actual_shots_on_target,
-- corners
lp.predicted_corners AS llm_corners,
op.predicted_corners AS openai_corners,
mts.corners AS actual_corners,
-- fouls
lp.predicted_fouls AS llm_fouls,
op.predicted_fouls AS openai_fouls,
mts.fouls AS actual_fouls,
-- yellow cards
lp.predicted_yellow_cards AS llm_yellow_cards,
op.predicted_yellow_cards AS openai_yellow_cards,
mts.yellow_cards AS actual_yellow_cards,
op.confidence AS openai_confidence,
op.model AS openai_model
FROM bookie.match m
JOIN bookie.team t
ON t.team_id IN (m.home_team_id, m.away_team_id)
JOIN bookie.team opp
ON opp.team_id = CASE WHEN m.home_team_id = t.team_id THEN m.away_team_id ELSE m.home_team_id END
LEFT JOIN bookie.match_prediction lp
ON lp.match_id = m.match_id AND lp.team_id = t.team_id
LEFT JOIN bookie.v_match_prediction_openai_latest op
ON op.match_id = m.match_id AND op.team_id = t.team_id
LEFT JOIN bookie.match_team_stats mts
ON mts.match_id = m.match_id AND mts.team_id = t.team_id
WHERE lp.prediction_id IS NOT NULL OR op.prediction_id IS NOT NULL;
-- ---------------------------------------------------------------------
-- 4) Accuracy summary: mean absolute error of each model vs actual,
-- per metric, over finished matches only. Lower = better.
-- ---------------------------------------------------------------------
CREATE OR REPLACE VIEW bookie.v_prediction_accuracy AS
WITH c AS (
SELECT * FROM bookie.v_prediction_comparison WHERE status = 'finished'
)
SELECT metric,
ROUND(AVG(ABS(llm_pred - actual)), 3) AS llm_mae,
ROUND(AVG(ABS(openai_pred - actual)), 3) AS openai_mae,
COUNT(*) FILTER (WHERE llm_pred IS NOT NULL AND actual IS NOT NULL) AS llm_samples,
COUNT(*) FILTER (WHERE openai_pred IS NOT NULL AND actual IS NOT NULL) AS openai_samples
FROM (
SELECT 'goals' AS metric, llm_goals AS llm_pred, openai_goals AS openai_pred, actual_goals AS actual FROM c
UNION ALL
SELECT 'shots_on_target', llm_shots_on_target, openai_shots_on_target, actual_shots_on_target FROM c
UNION ALL
SELECT 'corners', llm_corners, openai_corners, actual_corners FROM c
UNION ALL
SELECT 'fouls', llm_fouls, openai_fouls, actual_fouls FROM c
UNION ALL
SELECT 'yellow_cards', llm_yellow_cards, openai_yellow_cards, actual_yellow_cards FROM c
) x
GROUP BY metric
ORDER BY metric;

View File

@@ -0,0 +1,227 @@
-- =====================================================================
-- Inspection views for Dixon-Coles model + betting odds tables (bookie)
-- Assumes tables already exist. Safe to re-run (CREATE OR REPLACE).
--
-- psql -U <user> -d <db> -f sql/model_and_odds_views.sql
-- =====================================================================
-- Predictions with match & team context
CREATE OR REPLACE VIEW bookie.v_match_prediction_detail AS
SELECT
mp.prediction_id,
mp.match_id,
m.kickoff_at,
m.status,
ht.name AS home_team,
at.name AS away_team,
l.name AS league,
l.country,
t.name AS team_name,
(m.home_team_id = t.team_id) AS is_home,
mp.predicted_goals,
mp.predicted_shots_total,
mp.predicted_shots_on_target,
mp.predicted_corners,
mp.predicted_fouls,
mp.predicted_yellow_cards,
mp.model_trained_at,
mp.half_life_days,
mp.predicted_at
FROM bookie.match_prediction mp
JOIN bookie.match m ON m.match_id = mp.match_id
JOIN bookie.team t ON t.team_id = mp.team_id
JOIN bookie.team ht ON ht.team_id = m.home_team_id
JOIN bookie.team at ON at.team_id = m.away_team_id
JOIN bookie.matchday md ON md.matchday_id = m.matchday_id
JOIN bookie.season s ON s.season_id = md.season_id
JOIN bookie.league l ON l.league_id = s.league_id
ORDER BY m.kickoff_at DESC, mp.match_id, is_home DESC;
-- League Dixon-Coles parameters
CREATE OR REPLACE VIEW bookie.v_league_model_param_detail AS
SELECT
l.league_id,
l.name AS league_name,
l.country,
l.tier_level,
p.home_advantage,
p.rho,
p.avg_home_goals,
p.avg_away_goals,
p.matches_used,
p.half_life_days,
p.trained_at,
(SELECT COUNT(*) FROM bookie.team_strength ts WHERE ts.league_id = l.league_id) AS team_strength_count
FROM bookie.league l
LEFT JOIN bookie.league_model_param p ON p.league_id = l.league_id
ORDER BY l.country, l.name;
-- Team attack/defense with readable multipliers
CREATE OR REPLACE VIEW bookie.v_team_strength_detail AS
SELECT
ts.team_strength_id,
ts.team_id,
t.name AS team_name,
ts.league_id,
l.name AS league_name,
l.country,
ts.log_attack,
ts.log_defense,
ROUND(EXP(ts.log_attack)::numeric, 4) AS attack_factor,
ROUND(EXP(ts.log_defense)::numeric, 4) AS defense_factor,
ts.matches_used,
ts.trained_at
FROM bookie.team_strength ts
JOIN bookie.team t ON t.team_id = ts.team_id
JOIN bookie.league l ON l.league_id = ts.league_id
ORDER BY l.country, l.name, attack_factor DESC;
-- 1X2 odds per bookmaker with match context
CREATE OR REPLACE VIEW bookie.v_match_odds_detail AS
SELECT
mo.match_odds_id,
mo.match_id,
m.kickoff_at,
m.status,
ht.name AS home_team,
at.name AS away_team,
l.name AS league,
mo.bookmaker,
mo.home_odds,
mo.draw_odds,
mo.away_odds,
ROUND((1 / mo.home_odds)::numeric, 4) AS home_implied,
ROUND((1 / mo.draw_odds)::numeric, 4) AS draw_implied,
ROUND((1 / mo.away_odds)::numeric, 4) AS away_implied,
mo.total_line,
mo.over_odds,
mo.under_odds,
CASE WHEN mo.over_odds IS NOT NULL THEN ROUND((1 / mo.over_odds)::numeric, 4) END AS over_implied,
CASE WHEN mo.under_odds IS NOT NULL THEN ROUND((1 / mo.under_odds)::numeric, 4) END AS under_implied,
mo.fetched_at
FROM bookie.match_odds mo
JOIN bookie.match m ON m.match_id = mo.match_id
JOIN bookie.team ht ON ht.team_id = m.home_team_id
JOIN bookie.team at ON at.team_id = m.away_team_id
JOIN bookie.matchday md ON md.matchday_id = m.matchday_id
JOIN bookie.season s ON s.season_id = md.season_id
JOIN bookie.league l ON l.league_id = s.league_id
ORDER BY m.kickoff_at DESC, mo.match_id, mo.bookmaker;
-- Averaged 1X2 odds per match (what predict uses for market blending)
CREATE OR REPLACE VIEW bookie.v_match_odds_consensus AS
SELECT
mo.match_id,
m.kickoff_at,
m.status,
ht.name AS home_team,
at.name AS away_team,
l.name AS league,
COUNT(*) AS bookmaker_count,
ROUND(AVG(mo.home_odds)::numeric, 3) AS avg_home_odds,
ROUND(AVG(mo.draw_odds)::numeric, 3) AS avg_draw_odds,
ROUND(AVG(mo.away_odds)::numeric, 3) AS avg_away_odds,
ROUND((1 / AVG(mo.home_odds))::numeric, 4) AS home_implied,
ROUND((1 / AVG(mo.draw_odds))::numeric, 4) AS draw_implied,
ROUND((1 / AVG(mo.away_odds))::numeric, 4) AS away_implied,
ROUND(AVG(mo.total_line)::numeric, 2) AS avg_total_line,
ROUND(AVG(mo.over_odds)::numeric, 3) AS avg_over_odds,
ROUND(AVG(mo.under_odds)::numeric, 3) AS avg_under_odds,
CASE WHEN AVG(mo.over_odds) IS NOT NULL
THEN ROUND((1 / AVG(mo.over_odds))::numeric, 4) END AS over_implied,
CASE WHEN AVG(mo.under_odds) IS NOT NULL
THEN ROUND((1 / AVG(mo.under_odds))::numeric, 4) END AS under_implied,
COUNT(mo.total_line) AS totals_bookmaker_count,
MAX(mo.fetched_at) AS latest_fetched_at
FROM bookie.match_odds mo
JOIN bookie.match m ON m.match_id = mo.match_id
JOIN bookie.team ht ON ht.team_id = m.home_team_id
JOIN bookie.team at ON at.team_id = m.away_team_id
JOIN bookie.matchday md ON md.matchday_id = m.matchday_id
JOIN bookie.season s ON s.season_id = md.season_id
JOIN bookie.league l ON l.league_id = s.league_id
GROUP BY mo.match_id, m.kickoff_at, m.status, ht.name, at.name, l.name
ORDER BY m.kickoff_at DESC;
-- Team name aliases for odds API matching
CREATE OR REPLACE VIEW bookie.v_team_odds_alias_detail AS
SELECT
a.team_id,
t.name AS team_name,
a.odds_api_name,
(a.odds_api_name = t.name) AS exact_name_match
FROM bookie.team_odds_alias a
JOIN bookie.team t ON t.team_id = a.team_id
ORDER BY t.name;
-- Extra markets (corners/cards O/U)
CREATE OR REPLACE VIEW bookie.v_match_extra_odds_detail AS
SELECT
eo.match_extra_odds_id,
eo.match_id,
m.kickoff_at,
ht.name AS home_team,
at.name AS away_team,
eo.market,
eo.bookmaker,
eo.line,
eo.over_odds,
eo.under_odds,
eo.fetched_at
FROM bookie.match_extra_odds eo
JOIN bookie.match m ON m.match_id = eo.match_id
JOIN bookie.team ht ON ht.team_id = m.home_team_id
JOIN bookie.team at ON at.team_id = m.away_team_id
ORDER BY m.kickoff_at DESC, eo.match_id, eo.market, eo.bookmaker;
-- One-row-per-match snapshot for pre-match review (model + odds + prediction coverage)
CREATE OR REPLACE VIEW bookie.v_prematch_model_snapshot AS
SELECT
m.match_id,
m.kickoff_at,
m.status,
ht.name AS home_team,
at.name AS away_team,
l.league_id,
l.name AS league,
l.country,
s.name AS season,
lmp.home_advantage,
lmp.rho,
lmp.avg_home_goals,
lmp.avg_away_goals,
lmp.matches_used AS model_matches_used,
lmp.trained_at AS model_trained_at,
hs.log_attack AS home_log_attack,
hs.log_defense AS home_log_defense,
aws.log_attack AS away_log_attack,
aws.log_defense AS away_log_defense,
oc.bookmaker_count,
oc.avg_home_odds,
oc.avg_draw_odds,
oc.avg_away_odds,
oc.home_implied,
oc.draw_implied,
oc.away_implied,
oc.avg_total_line,
oc.avg_over_odds,
oc.avg_under_odds,
oc.over_implied,
oc.under_implied,
hp.predicted_goals AS home_pred_goals,
ap.predicted_goals AS away_pred_goals,
hp.predicted_at AS home_predicted_at,
ap.predicted_at AS away_predicted_at
FROM bookie.match m
JOIN bookie.team ht ON ht.team_id = m.home_team_id
JOIN bookie.team at ON at.team_id = m.away_team_id
JOIN bookie.matchday md ON md.matchday_id = m.matchday_id
JOIN bookie.season s ON s.season_id = md.season_id
JOIN bookie.league l ON l.league_id = s.league_id
LEFT JOIN bookie.league_model_param lmp ON lmp.league_id = l.league_id
LEFT JOIN bookie.team_strength hs ON hs.team_id = m.home_team_id AND hs.league_id = l.league_id
LEFT JOIN bookie.team_strength aws ON aws.team_id = m.away_team_id AND aws.league_id = l.league_id
LEFT JOIN bookie.v_match_odds_consensus oc ON oc.match_id = m.match_id
LEFT JOIN bookie.match_prediction hp ON hp.match_id = m.match_id AND hp.team_id = m.home_team_id
LEFT JOIN bookie.match_prediction ap ON ap.match_id = m.match_id AND ap.team_id = m.away_team_id
ORDER BY m.kickoff_at DESC;