haunt-fm/src/haunt_fm/api/feedback.py

from fastapi import APIRouter, Depends, HTTPException, Query
from pydantic import BaseModel
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession
from sqlalchemy.orm import joinedload

from haunt_fm.db import get_session
from haunt_fm.models.track import FeedbackEvent, Track
from haunt_fm.services.feedback import compute_contextual_score, record_feedback

router = APIRouter(prefix="/api/feedback")


class FeedbackRequest(BaseModel):
    playlist_id: int
    track_id: int
    signal: str


class ScoreRequest(BaseModel):
    track_id: int
    vibe_text: str


@router.post("")
async def submit_feedback(req: FeedbackRequest, session: AsyncSession = Depends(get_session)):
    """Submit feedback for a track in the context of a playlist's vibe."""
    try:
        event = await record_feedback(session, req.playlist_id, req.track_id, req.signal)
    except ValueError as e:
        raise HTTPException(status_code=400, detail=str(e))

    return {
        "id": event.id,
        "playlist_id": event.playlist_id,
        "track_id": event.track_id,
        "signal": event.signal,
        "signal_weight": event.signal_weight,
        "vibe_text": event.vibe_text,
        "created_at": event.created_at.isoformat(),
    }


@router.post("/score")
async def get_score(req: ScoreRequest, session: AsyncSession = Depends(get_session)):
    """Compute the contextual feedback score for a track given a vibe description."""
    from haunt_fm.services.embedding import embed_text, is_model_loaded, load_model

    if not is_model_loaded():
        load_model()
    vibe_embedding = embed_text(req.vibe_text)

    result = await session.execute(
        select(FeedbackEvent).where(FeedbackEvent.track_id == req.track_id)
    )
    events = list(result.scalars().all())

    score = compute_contextual_score(events, vibe_embedding)

    # Build breakdown of contributing events
    import numpy as np

    breakdown = []
    for event in events:
        event_emb = np.array(event.vibe_embedding, dtype=np.float32)
        event_norm = np.linalg.norm(event_emb)
        vibe_norm = np.linalg.norm(vibe_embedding)
        if event_norm > 0 and vibe_norm > 0:
            cos_sim = float(np.dot(event_emb / event_norm, vibe_embedding / vibe_norm))
        else:
            cos_sim = 0.0

        breakdown.append({
            "id": event.id,
            "signal": event.signal,
            "signal_weight": event.signal_weight,
            "vibe_text": event.vibe_text,
            "cosine_similarity": round(cos_sim, 4),
            "created_at": event.created_at.isoformat(),
        })

    return {
        "track_id": req.track_id,
        "vibe_text": req.vibe_text,
        "contextual_score": round(score, 4),
        "event_count": len(events),
        "breakdown": breakdown,
    }


@router.get("/history")
async def get_history(
    limit: int = Query(default=50, ge=1, le=200),
    track_id: int | None = Query(default=None),
    session: AsyncSession = Depends(get_session),
):
    """Get recent feedback events, optionally filtered by track."""
    query = (
        select(FeedbackEvent)
        .options(joinedload(FeedbackEvent.track))
        .order_by(FeedbackEvent.created_at.desc())
        .limit(limit)
    )
    if track_id is not None:
        query = query.where(FeedbackEvent.track_id == track_id)

    result = await session.execute(query)
    events = result.scalars().unique().all()

    return {
        "events": [
            {
                "id": e.id,
                "playlist_id": e.playlist_id,
                "track_id": e.track_id,
                "artist": e.track.artist,
                "title": e.track.title,
                "signal": e.signal,
                "signal_weight": e.signal_weight,
                "vibe_text": e.vibe_text,
                "created_at": e.created_at.isoformat(),
            }
            for e in events
        ],
        "count": len(events),
    }