fde-worldcup-data-worker / scripts /worldcup /generate-score-analyses.mjs
wihaha's picture
fix: use worldcup match 90 minute score columns
4eb2847
Raw
History Blame Contribute Delete
11.8 kB
import crypto from 'node:crypto';
import { loadLocalEnv } from '../gaokao/lib/env.mjs';
import { withDb } from '../gaokao/lib/db.mjs';
loadLocalEnv();
const modelName = process.env.REASONING_MODEL || 'gemini-3.5-flash';
const maxMatches = Number(process.env.SCORE_ANALYSIS_MAX_MATCHES || 8);
function stableHash(value) {
return crypto.createHash('sha256').update(JSON.stringify(value)).digest('hex').slice(0, 24);
}
function extractJson(text) {
const fenced = text.match(/```json\s*([\s\S]*?)```/i);
const raw = fenced?.[1] || text;
const start = raw.indexOf('{');
const end = raw.lastIndexOf('}');
if (start < 0 || end < start) throw new Error('score analysis response did not include JSON');
return JSON.parse(raw.slice(start, end + 1));
}
async function ensureTable(pool) {
await pool.query(`
create table if not exists worldcup_score_analyses (
match_id text primary key references worldcup_matches(id),
status text not null,
model text,
predicted_score text,
score_probabilities jsonb not null default '[]'::jsonb,
summary_zh text,
reasoning_md text,
basis jsonb not null default '{}'::jsonb,
input_fingerprint text,
error_message text,
created_at timestamptz not null default now(),
updated_at timestamptz not null default now()
)
`);
}
async function callGemini(systemPrompt, userPrompt) {
const apiKey = process.env.VECTORENGINE_GEMINI_KEY || process.env.VECTORENGINE_API_KEY;
if (!apiKey) throw new Error('VECTORENGINE_GEMINI_KEY is not configured');
const apiBase = (process.env.VECTORENGINE_API_BASE || 'https://api.vectorengine.cn/v1').replace(/\/$/, '');
const response = await fetch(`${apiBase}/chat/completions`, {
method: 'POST',
headers: {
Authorization: `Bearer ${apiKey}`,
'Content-Type': 'application/json',
},
body: JSON.stringify({
model: modelName,
temperature: Number(process.env.SCORE_ANALYSIS_TEMPERATURE || 0.35),
messages: [
{ role: 'system', content: systemPrompt },
{ role: 'user', content: userPrompt },
],
}),
signal: AbortSignal.timeout(Number(process.env.SCORE_ANALYSIS_TIMEOUT_MS || 90000)),
});
if (!response.ok) {
const raw = await response.text().catch(() => '');
throw new Error(raw || `score analysis request failed: ${response.status}`);
}
const data = await response.json();
return data?.choices?.[0]?.message?.content || '';
}
async function getEligibleMatches(pool) {
const result = await pool.query(`
select
m.id,
m.kickoff_utc,
ht.name_zh as home_name_zh,
at.name_zh as away_name_zh,
count(distinct o.id) as odds_count,
count(distinct w.id) as weather_count
from worldcup_matches m
join worldcup_teams ht on ht.id = m.home_team_id
join worldcup_teams at on at.id = m.away_team_id
join worldcup_weather_snapshots w on w.match_id = m.id
join worldcup_market_odds_snapshots o on o.match_id = m.id
and o.market_key = 'h2h'
and o.home_odds is not null
and o.draw_odds is not null
and o.away_odds is not null
left join worldcup_score_analyses a on a.match_id = m.id and a.status = 'success'
where m.status in ('scheduled', 'active')
and a.match_id is null
group by m.id, m.kickoff_utc, ht.name_zh, at.name_zh
order by m.kickoff_utc
limit $1
`, [maxMatches]);
return result.rows;
}
async function getContext(pool, matchId) {
const matchRes = await pool.query(`
select
m.id,
m.stage,
m.round,
m.kickoff_utc,
m.home_team_id,
m.away_team_id,
ht.name_zh as home_name_zh,
ht.name_en as home_name_en,
at.name_zh as away_name_zh,
at.name_en as away_name_en,
v.name as venue_name,
v.city as venue_city,
v.country as venue_country
from worldcup_matches m
left join worldcup_teams ht on ht.id = m.home_team_id
left join worldcup_teams at on at.id = m.away_team_id
left join worldcup_venues v on v.id = m.venue_id
where m.id = $1
`, [matchId]);
if (!matchRes.rows[0]) throw new Error(`Match not found: ${matchId}`);
const match = matchRes.rows[0];
const rankingsRes = await pool.query(`
with ranked as (
select team_id, ranking_type, rank, rating,
row_number() over (partition by team_id, ranking_type order by ranking_date desc) rn
from worldcup_team_rankings
where team_id in ($1, $2)
)
select team_id, ranking_type, rank, rating
from ranked
where rn = 1
`, [match.home_team_id, match.away_team_id]);
const formRes = await pool.query(`
with ranked as (
select team_id, match_date, opponent_name_raw, competition, result, goals_for, goals_against, opponent_elo,
row_number() over (partition by team_id order by match_date desc) rn
from worldcup_team_form
where team_id in ($1, $2)
)
select *
from ranked
where rn <= 10
order by team_id, match_date desc
`, [match.home_team_id, match.away_team_id]);
const weatherRes = await pool.query(`
select forecast_time, temperature_c, apparent_temperature_c, humidity_pct,
precipitation_probability_pct, precipitation_mm, wind_speed_kmh, wind_gusts_kmh, weather_code
from worldcup_weather_snapshots
where match_id = $1
order by snapshot_time desc
limit 1
`, [matchId]);
const oddsRes = await pool.query(`
with ranked as (
select bookmaker_key, bookmaker_title, market_key, market_title,
home_odds, draw_odds, away_odds, last_update,
row_number() over (
partition by bookmaker_key, market_key
order by coalesce(last_update, snapshot_time) desc, snapshot_time desc
) rn
from worldcup_market_odds_snapshots
where match_id = $1
and market_key = 'h2h'
and home_odds is not null
and draw_odds is not null
and away_odds is not null
)
select *
from ranked
where rn = 1
order by bookmaker_title
`, [matchId]);
const tournamentResultsRes = await pool.query(`
select
m.id,
m.stage,
m.round,
m.kickoff_utc,
m.home_team_id,
m.away_team_id,
ht.name_zh as home_name_zh,
at.name_zh as away_name_zh,
m.home_score_90,
m.away_score_90,
case
when m.home_score_90 > m.away_score_90 then m.home_team_id
when m.away_score_90 > m.home_score_90 then m.away_team_id
else 'draw'
end as result_side
from worldcup_matches m
left join worldcup_teams ht on ht.id = m.home_team_id
left join worldcup_teams at on at.id = m.away_team_id
where m.status = 'finished'
and m.home_score_90 is not null
and m.away_score_90 is not null
and (
m.home_team_id in ($1, $2)
or m.away_team_id in ($1, $2)
)
order by m.kickoff_utc desc
limit 12
`, [match.home_team_id, match.away_team_id]);
return {
match,
rankings: rankingsRes.rows,
recent_form: formRes.rows,
weather: weatherRes.rows[0] || null,
odds: oddsRes.rows,
tournament_results: tournamentResultsRes.rows,
};
}
async function saveAnalysis(pool, matchId, payload) {
await pool.query(`
insert into worldcup_score_analyses (
match_id, status, model, predicted_score, score_probabilities,
summary_zh, reasoning_md, basis, input_fingerprint, error_message, updated_at
) values ($1,$2,$3,$4,$5::jsonb,$6,$7,$8::jsonb,$9,$10,now())
on conflict (match_id) do update set
status = excluded.status,
model = excluded.model,
predicted_score = excluded.predicted_score,
score_probabilities = excluded.score_probabilities,
summary_zh = excluded.summary_zh,
reasoning_md = excluded.reasoning_md,
basis = excluded.basis,
input_fingerprint = excluded.input_fingerprint,
error_message = excluded.error_message,
updated_at = now()
`, [
matchId,
payload.status,
payload.model || null,
payload.predicted_score || null,
JSON.stringify(payload.score_probabilities || []),
payload.summary_zh || null,
payload.reasoning_md || null,
JSON.stringify(payload.basis || {}),
payload.input_fingerprint || null,
payload.error_message || null,
]);
}
function buildPrompt(context) {
const systemPrompt = `你是世界杯预测分析师。你必须基于给定的结构化数据,输出比分概率,而不是泛泛聊天。
要求:
- 使用中文。
- 结合 Elo/FIFA 排名、近 10 场状态、本届已完赛表现、天气、赔率盘口。
- 不要声称掌握未提供的首发或伤病。
- 如果本届已完赛数据与长期实力数据矛盾,要明确说明哪一项权重更高以及原因。
- 盘口只作为市场共识补充,不要机械等同于最终概率。
- 给出 5 个最可能比分及概率,概率总和不必为 100%,但每个概率必须合理。
- 推理要比一句话更充分,必须包含主要证据、反向风险和比分路径。
- 输出必须是一个 JSON 对象,不要 markdown,不要额外文字。
JSON schema:
{
"predicted_score": "2-1",
"score_probabilities": [
{"score": "2-1", "probability": 0.14, "label_zh": "主队小胜"}
],
"summary_zh": "一句话结论",
"reasoning_md": "Markdown 格式的推理依据,建议包含:结论、模型基础、本届比赛表现、盘口信号、天气影响、比分路径、风险因素。",
"basis": {
"main_factors": ["Elo/FIFA差距", "本届比赛表现", "市场赔率", "天气"],
"data_quality": "complete"
}
}`;
return {
systemPrompt,
userPrompt: JSON.stringify(context, null, 2),
};
}
async function main() {
const apiKey = process.env.VECTORENGINE_GEMINI_KEY || process.env.VECTORENGINE_API_KEY;
if (!apiKey) {
console.log('[score-analysis] VECTORENGINE_GEMINI_KEY is not configured; skipping score analysis');
return;
}
await withDb(async (pool) => {
await ensureTable(pool);
const matches = await getEligibleMatches(pool);
console.log(`[score-analysis] eligible=${matches.length} max=${maxMatches}`);
let success = 0;
let failed = 0;
for (const match of matches) {
try {
const context = await getContext(pool, match.id);
if (!context.weather || !context.odds.length) {
console.log(`[score-analysis] skip ${match.id} missing complete weather/odds`);
continue;
}
const fingerprint = stableHash(context);
const { systemPrompt, userPrompt } = buildPrompt(context);
const raw = await callGemini(systemPrompt, userPrompt);
const parsed = extractJson(raw);
await saveAnalysis(pool, match.id, {
status: 'success',
model: modelName,
predicted_score: parsed.predicted_score,
score_probabilities: parsed.score_probabilities || [],
summary_zh: parsed.summary_zh || '',
reasoning_md: parsed.reasoning_md || '',
basis: parsed.basis || {},
input_fingerprint: fingerprint,
});
success += 1;
console.log(`[score-analysis] ${match.id} ${match.home_name_zh} vs ${match.away_name_zh} predicted=${parsed.predicted_score}`);
} catch (error) {
failed += 1;
await saveAnalysis(pool, match.id, {
status: 'failed',
model: modelName,
error_message: error.message || String(error),
}).catch(() => {});
console.warn(`[score-analysis] ${match.id} failed: ${error.message || error}`);
}
}
console.log(`[score-analysis] complete success=${success} failed=${failed}`);
});
}
await main();