alvis/oO
1
0
Fork 0
Code Issues 34 Pull Requests Actions Packages Projects Releases Wiki Activity

feat(bandit): promote egreedy-v2 (D=12, profile features) as active policy (#99)

Browse Source 
Offline sim gate passed — egreedy-v2 mean reward −0.629 vs egreedy-v1 −0.642
(5 users × 20 rounds, rule judge, seed 42). v2 wins 3/5 personas.

- recommender.ts: switch remotePolicy() to /score/egreedy/v2
- recommender.ts: switch sendRewardWithRetry() to /reward/egreedy/v2 with
  profile_features payload so the ridge update uses the full D=12 vector
- recommender.ts: re-fetch profile at feedback time (TTL-cached, near-instant)
- ADR-0012: status Accepted → Promoted, promotion record appended

Shadow entry egreedy-v2-shadow kept in registry (active: false) for rollback.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
alvis
2026-04-26 03:08:28 +00:00
parent cba3f1a184
commit 7281af83a4
2 changed files with 29 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
services/api/src/routes/recommender.ts
View File

@@ -47,7 +47,8 @@ export const _clearCandidateCacheForTests = () => {
// Shadow-policy registry
// ---------------------------------------------------------------------------
const shadowPolicies = new Map<string, { active: boolean }>([
// egreedy-v2 (D=12, profile features) — disabled until sim gate per ADR-0012
// egreedy-v2 promoted to active policy (ADR-0012). Shadow entry kept for
// rollback toggle; leave disabled in normal operation.
['egreedy-v2-shadow', { active: false }],
]);
@@ -101,9 +102,9 @@ async function remotePolicy(
profile_features: profile,
};
// Active policy: egreedy-v1 (selected over linucb-v1 after offline sim — ADR-0007)
// Active policy: egreedy-v2 (promoted from shadow after offline sim — ADR-0012)
try {
const res = await fetch(`${config.ML_SERVING_URL}/score/egreedy`, {
const res = await fetch(`${config.ML_SERVING_URL}/score/egreedy/v2`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify(body),
@@ -111,7 +112,7 @@ async function remotePolicy(
});
if (!res.ok) return null;
const data = (await res.json()) as { tip_id: string; score: number };
return { tipId: data.tip_id, score: data.score, policy: 'egreedy-v1' };
return { tipId: data.tip_id, score: data.score, policy: 'egreedy-v2' };
} catch {
return null;
}
@@ -371,6 +372,7 @@ async function sendRewardWithRetry(
tipId: string,
reward: number,
features: TipCandidate['features'],
profile: Profile,
): Promise<void> {
const body = JSON.stringify({
user_id: userId,
@@ -378,11 +380,12 @@ async function sendRewardWithRetry(
reward,
features,
day_of_week: new Date().getDay(),
profile_features: profile,
});
for (let attempt = 1; attempt <= 3; attempt++) {
try {
const res = await fetch(`${config.ML_SERVING_URL}/reward/egreedy`, {
const res = await fetch(`${config.ML_SERVING_URL}/reward/egreedy/v2`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body,
@@ -463,7 +466,9 @@ router.post('/tip/:id/feedback', requireAuth, async (req: AuthenticatedRequest,
});
if (candidate) {
sendRewardWithRetry(req.userId!, tipId, reward, candidate.features);
// Re-fetch profile for the v2 ridge update; TTL cache makes this near-instant.
const profile = await getProfile(req.userId!);
sendRewardWithRetry(req.userId!, tipId, reward, candidate.features, profile);
}
// Delegate action to the owning signal source (e.g. mark done in Todoist)