app.py

# app.py
from datetime import datetime, timedelta
from io import BytesIO
import base64, json
import random  # <-- for random tip selection

from flask import Flask, request, jsonify, render_template
from sqlalchemy import text, func
from PIL import Image
import numpy as np

# Torch
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import models
from torchvision.models import EfficientNet_B0_Weights

# Miscellaneous imports from misc/
from misc.extensions import db, login_manager
from misc.models import User, ClassificationLog, normalize_label
from misc.auth import auth_bp, login_required, current_user  # re-exported from auth
from misc.auth import auth_bp, api_logout as bp_api_logout

from misc.policy_engine import decide_action

# ---------------- App & Config ----------------
app = Flask(__name__)
app.config["SECRET_KEY"] = "change-me"               # set via env var in prod
app.config["SQLALCHEMY_DATABASE_URI"] = "sqlite:///recycloai.db"
app.config["SQLALCHEMY_TRACK_MODIFICATIONS"] = False

db.init_app(app)
login_manager.init_app(app)
login_manager.login_view = "auth.login"  # type: ignore[assignment]

# ---------------- SQLite schema repair ----------------
def _table_exists(conn, name: str) -> bool:
    row = conn.execute(text(
        "SELECT name FROM sqlite_master WHERE type='table' AND name=:n"
    ), {"n": name}).fetchone()
    return row is not None

def _cols(conn, name: str) -> set[str]:
    rows = conn.execute(text(f"PRAGMA table_info({name})")).fetchall()
    return {r[1] for r in rows}  # r[1] = column name

def ensure_sqlite_schema(app: Flask):
    """
    Make the live recycloai.db compatible with our models without losing data.
    - If a legacy 'user' table exists, rename to 'users' (if 'users' missing).
    - Create missing tables.
    - Add any missing columns on existing tables.
    """
    with app.app_context():
        with db.engine.begin() as conn:
            # 1) Rename legacy table 'user' -> 'users' if needed
            has_users = _table_exists(conn, "users")
            has_user  = _table_exists(conn, "user")
            if has_user and not has_users:
                conn.execute(text("ALTER TABLE user RENAME TO users"))

        # 2) Create missing tables (no-op if they already exist)
        db.create_all()

        with db.engine.begin() as conn:
            # 3) Add missing columns on 'users' (or legacy 'user' if both exist)
            target_user_table = "users" if _table_exists(conn, "users") else ("user" if _table_exists(conn, "user") else None)
            if target_user_table:
                ucols = _cols(conn, target_user_table)
                if "email" not in ucols:
                    conn.execute(text(f"ALTER TABLE {target_user_table} ADD COLUMN email TEXT"))
                if "name" not in ucols:
                    conn.execute(text(f"ALTER TABLE {target_user_table} ADD COLUMN name TEXT"))
                if "password_hash" not in ucols:
                    conn.execute(text(f"ALTER TABLE {target_user_table} ADD COLUMN password_hash TEXT"))

            # 4) Ensure 'classification_logs' table exists and has all columns
            if not _table_exists(conn, "classification_logs"):
                # create_all should have created it; if not, force-create
                conn.execute(text("""
                    CREATE TABLE IF NOT EXISTS classification_logs (
                        id INTEGER PRIMARY KEY,
                        user_id INTEGER,
                        label TEXT,
                        confidence REAL,
                        city TEXT,
                        created_at DATETIME
                    )
                """))

            lcols = _cols(conn, "classification_logs")
            if "user_id" not in lcols:
                conn.execute(text("ALTER TABLE classification_logs ADD COLUMN user_id INTEGER"))
            if "label" not in lcols:
                conn.execute(text("ALTER TABLE classification_logs ADD COLUMN label TEXT"))
            if "confidence" not in lcols:
                conn.execute(text("ALTER TABLE classification_logs ADD COLUMN confidence REAL"))
            if "city" not in lcols:
                conn.execute(text("ALTER TABLE classification_logs ADD COLUMN city TEXT"))
            if "created_at" not in lcols:
                conn.execute(text("ALTER TABLE classification_logs ADD COLUMN created_at DATETIME"))

# ---------------- Model / Inference setup ----------------
STATE_PATH = "best_efficientnet_model.pth"
CLASS_NAMES_PATH = "artifacts/class_names.json"
NUM_CLASSES_FALLBACK = 6
THRESH = 0.75

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
_state = torch.load(STATE_PATH, map_location="cpu")

def _infer_num_classes(state_dict: dict) -> int | None:
    for k, v in state_dict.items():
        if k.endswith("classifier.1.weight") and hasattr(v, "shape"):
            return int(v.shape[0])
    for k, v in state_dict.items():
        if k.endswith("classifier.1.bias") and hasattr(v, "shape"):
            return int(v.shape[0])
    return None

num_classes = _infer_num_classes(_state) or NUM_CLASSES_FALLBACK

_model = models.efficientnet_b0(weights=EfficientNet_B0_Weights.DEFAULT)
# grab in_features
def _in_features(classifier: nn.Module) -> int:
    if isinstance(classifier, nn.Linear):
        return int(classifier.in_features)
    if isinstance(classifier, nn.Sequential):
        for mod in reversed(classifier):
            if isinstance(mod, nn.Linear):
                return int(mod.in_features)
    return 1280

in_features = _in_features(_model.classifier)
_model.classifier = nn.Sequential(nn.Dropout(0.2), nn.Linear(in_features, num_classes))
_missing, _unexpected = _model.load_state_dict(_state, strict=False)
_model.eval()
model = _model.to(device)

def _load_class_names(path: str) -> list[str]:
    try:
        with open(path, "r") as f:
            classes = json.load(f)
        if isinstance(classes, list) and all(isinstance(x, str) for x in classes):
            return classes
    except Exception:
        pass
    return ["Cardboard", "Glass", "Metal", "Paper", "Plastic", "Trash"]

CLASS_NAMES = _load_class_names(CLASS_NAMES_PATH)

def prepare_image(img: Image.Image) -> torch.Tensor:
    img = img.convert('RGB').resize((224, 224))
    arr = np.array(img).astype(np.float32) / 255.0
    mean = np.array([0.485, 0.456, 0.406], dtype=np.float32)
    std  = np.array([0.229, 0.224, 0.225], dtype=np.float32)
    arr = (arr - mean) / std
    arr = np.transpose(arr, (2, 0, 1))
    return torch.from_numpy(arr).unsqueeze(0).to(device)

# ---------------- Tips (3 per class/action) + random selection ----------------
TIPS = {
    "Cardboard": [
        "Most recycling bins accept clean cardboard — look for the blue or green bin with the recycling logo.",
        "If you have too much, take flattened boxes to a local recycling drop-off center or grocery store collection point.",
        "Wet or food-soiled cardboard (like greasy pizza boxes) usually goes in the trash or compost, depending on city rules."
    ],
    "Glass": [
        "Recycle bottles and jars in your household recycling bin if your city accepts glass — check your city’s website or A-Z waste guide.",
        "If curbside glass recycling isn’t available, bring clean bottles and jars to a community recycling drop-off or bottle depot.",
        "Broken glass should go in the trash (wrapped safely) — it’s not recyclable curbside in most places."
    ],
    "Metal": [
        "Empty and rinse cans — then put them in your curbside recycling bin if your area accepts metal.",
        "If you have scrap metal (like wires or tools), take it to a local scrap yard or recycling center — many pay for metal.",
        "Clean aluminum foil can go in recycling if your city allows it; otherwise, collect and drop it off with metals at a facility."
    ],
    "Paper": [
        "Most clean paper (like office paper, mail, newspapers) goes in your curbside recycling bin.",
        "Check with your city if they accept shredded paper — some want it bagged, others ask for drop-off only.",
        "Glossy, waxy, or dirty paper (like takeout boxes) often goes in the trash or compost instead."
    ],
    "Plastic": [
        "Look for a recycling symbol (♻️) with numbers 1–2 — these are accepted in most curbside programs.",
        "If your plastic is soft or flexible (like bags or wraps), take it to store collection bins labeled ‘Plastic Film Recycling’.",
        "When unsure, search your local recycling guide by item name (e.g., 'plastic cup') — it’ll tell you where to drop it off."
    ],
    "Trash": [
        "If it can’t go in your recycling or compost, it belongs in the trash bin.",
        "Check your city’s disposal guide — some items (like batteries or electronics) need special drop-off locations.",
        "When possible, look for reuse options — donation centers, repair shops, or creative reuse stores often accept items."
    ],
    "Unsure": [
        "Try another angle, better lighting, or remove background clutter.",
        "Manually select a material or add attributes like ‘greasy/wet’.",
        "Check your local recycling guide for specific items."
    ],
}


ACTION_TIPS = {
    "Recyclable": [
        "Rinse/empty items and keep them dry to avoid contamination.",
        "Don’t bag recyclables—place them loose in the cart.",
        "If a piece is smaller than a credit card, it may not get captured."
    ],
    "Compost": [
        "Remove plastic liners or stickers; only food-soiled fiber belongs.",
        "Tear large pieces into smaller bits to speed up composting.",
        "No plastics, glass, or metal in organics—even if ‘biodegradable’."
    ],
    "Landfill": [
        "If it’s not accepted locally, place it in trash—don’t wish-cycle.",
        "Reduce and reuse where possible to cut landfill waste.",
        "Bundle messy trash to prevent leaks and pests."
    ],
    "Unsure": TIPS["Unsure"],  # reuse same three
}

def _pick_from_list(items: list[str]) -> str:
    """Pick a tip based on a random number from 1 to 3 (index 0..2)."""
    if not items:
        return "Check local recycling guidelines for your area."
    n = random.randint(1, 3)  # user request: choose a random number from 1..3
    # If the list has fewer than 3 items, wrap safely
    idx = (n - 1) % len(items)
    return items[idx]

def _tip_for(label: str | None, action: str | None) -> str:
    """
    Case-insensitive material tip lookup with random choice among 3 tips.
    Fallback to action-level tips, then to a generic default.
    """
    # Try material tips (case-insensitive)
    if label:
        key = label.strip()
        if key in TIPS:
            return _pick_from_list(TIPS[key])
        low = key.lower()
        for k in TIPS.keys():
            if k.lower() == low:
                return _pick_from_list(TIPS[k])
    # Fallback to action-level tips
    if action and action in ACTION_TIPS:
        return _pick_from_list(ACTION_TIPS[action])
    return "Check local recycling guidelines for your area."

# ---------------- Blueprints ----------------
app.register_blueprint(auth_bp)  # /login, /signup, /api/logout

# ---- add the alias RIGHT AFTER blueprint registration ----
@app.route("/api/logout", methods=["POST"], endpoint="api_logout")
def api_logout_alias():
    return bp_api_logout()

@app.route("/dashboard")
@login_required
def dashboard():
    return render_template("progress.html")

# ---------------- Pages ----------------
@app.route("/", endpoint="index")
def home():
    return render_template("home.html")  # or "index.html" if that's your file

# --- keep this public ---
@app.route("/charities")
def charities():
    return render_template("charities.html")

@app.route("/progress")
@login_required
def progress():
    return render_template("progress.html")

# ---------------- Health ----------------
@app.route("/health")
def health():
    return jsonify({
        "status": "ok",
        "device": str(device),
        "classes": CLASS_NAMES,
        "missing_keys": len(_missing),
        "unexpected_keys": len(_unexpected)
    })

# ---------------- Inference ----------------
@app.route("/process_image", methods=["POST"])
def process_image():
    data = request.get_json() or {}
    img_data = data.get("image_data")
    if not img_data:
        return jsonify({"error": "No image data provided."}), 400

    try:
        _, encoded = img_data.split(",", 1)
    except ValueError:
        encoded = img_data

    try:
        img_bytes = base64.b64decode(encoded)
        img = Image.open(BytesIO(img_bytes))
    except Exception as e:
        return jsonify({"error": f"Invalid image data: {e}"}), 400

    x = prepare_image(img)
    with torch.no_grad():
        logits = model(x)
        probs = F.softmax(logits, dim=1).cpu().numpy()[0]

    pred_idx = int(np.argmax(probs))
    confidence = float(probs[pred_idx])
    label = CLASS_NAMES[pred_idx] if 0 <= pred_idx < len(CLASS_NAMES) else f"Class_{pred_idx}"

    attrs = data.get("attrs") or {}
    user_city = data.get("city") or "default"

    # abstain
    if confidence < 0.75:
        action = "Unsure"
        resp = {
            "material": label,
            "action": action,
            "why": "Low confidence prediction. Try another angle or better light.",
            "confidence": confidence,
            "confidence_text": f"{confidence*100:.1f} % (low)",
            "tip": _tip_for("Unsure", "Unsure"),
            "abstained": True
        }
        try:
            if current_user.is_authenticated:
                db.session.add(ClassificationLog(
                    user_id=current_user.id,
                    label=normalize_label(action),
                    confidence=confidence,
                    city=user_city
                ))
                db.session.commit()
        except Exception:
            db.session.rollback()
        return jsonify(resp)

    # policy decision
    action, why = decide_action(label, attrs, user_city)

    # log per-user
    try:
        if current_user.is_authenticated:
            db.session.add(ClassificationLog(
                user_id=current_user.id,
                label=normalize_label(action),
                confidence=confidence,
                city=user_city
            ))
            db.session.commit()
    except Exception:
        db.session.rollback()

    return jsonify({
        "material": label,
        "action": action,
        "why": why,
        "confidence": confidence,
        "confidence_text": f"{confidence*100:.1f} % Confidence Score",
        "tip": _tip_for(label, action),
        "abstained": False
    })

# ---------------- Progress APIs ----------------
@app.route("/api/progress/summary", methods=["GET"])
@login_required
def api_progress_summary():
    # overall totals
    totals = {"Recyclable": 0, "Compost": 0, "Landfill": 0, "Unsure": 0, "Other": 0}
    rows = (db.session.query(ClassificationLog.label, func.count())
            .filter(ClassificationLog.user_id == current_user.id)
            .group_by(ClassificationLog.label).all())
    for label, cnt in rows:
        totals[normalize_label(label)] = cnt

    # last 14 days
    today = datetime.utcnow().date()
    since = today - timedelta(days=13)
    since_dt = datetime.combine(since, datetime.min.time())

    by_day = {(since + timedelta(days=i)).isoformat():
              {"Recyclable":0,"Compost":0,"Landfill":0,"Unsure":0,"Other":0}
              for i in range(14)}

    per = (db.session.query(
              text("date(created_at) as day"),
              ClassificationLog.label,
              func.count().label("cnt"))
           .filter(ClassificationLog.user_id == current_user.id)
           .filter(ClassificationLog.created_at >= since_dt)
           .group_by(text("day"), ClassificationLog.label)
           .all())

    for day, label, cnt in per:
        k = str(day)
        if k in by_day:
            by_day[k][normalize_label(label)] += cnt

    total = sum(totals.values())
    return jsonify({"ok": True, "total": total, "totals": totals, "per_day": by_day})

@app.route("/api/progress/logs", methods=["GET"])
@login_required
def api_progress_logs():
    limit = min(int(request.args.get("limit", 200)), 1000)
    logs = (ClassificationLog.query
            .filter_by(user_id=current_user.id)
            .order_by(ClassificationLog.created_at.desc())
            .limit(limit).all())
    return jsonify({
        "ok": True,
        "logs": [{
            "id": l.id,
            "ts": l.created_at.isoformat(),
            "label": l.label,
            "confidence": l.confidence,
            "city": l.city
        } for l in logs]
    })

@app.route("/api/logs", methods=["DELETE"])
@login_required
def api_clear_logs():
    ClassificationLog.query.filter_by(user_id=current_user.id).delete()
    db.session.commit()
    return jsonify({"ok": True})

# ---------------- Main ----------------
if __name__ == "__main__":
    ensure_sqlite_schema(app)   # repair/align existing DB
    app.run(debug=True)