ble-ai-localizer/app/train_mode.py

# app/train_mode.py
import os
import glob
import time
import math
import joblib
import numpy as np
import pandas as pd
from datetime import datetime
from sklearn.neighbors import KNeighborsClassifier, KNeighborsRegressor

# Import assoluti garantiti
from csv_config import load_gateway_features_csv
from logger_utils import log_msg as log

def run_train(settings, log_fn=None, target_files=None):
    """
    Esegue l'addestramento Hierarchical KNN su un set di file specifico (Campagna).
    """
    if log_fn is None:
        log_fn = log

    train_cfg = settings.get("train", {})
    knn_cfg = train_cfg.get("knn", {})
    
    # Parametri da config.yaml
    samples_dir = train_cfg.get("samples_dir", "/data/train/samples")
    gateways_csv = train_cfg.get("gateways_csv", "/data/config/gateway.csv")
    model_path = train_cfg.get("model_path", "/data/model/model.joblib")
    nan_fill = float(train_cfg.get("nan_fill", -110.0))
    
    k_val = int(knn_cfg.get("k", 5))
    weights = knn_cfg.get("weights", "distance")
    metric = knn_cfg.get("metric", "euclidean")

    log_fn(f"TRAIN: Caricamento gateway da {gateways_csv}")
    gws = load_gateway_features_csv(gateways_csv)
    gateways_order = [g.mac for g in gws]

    # Selezione file (Campagna specifica o globale)
    files = target_files if target_files else glob.glob(os.path.join(samples_dir, "*.csv"))
    if not files:
        raise RuntimeError("Nessun file CSV trovato per l'addestramento.")

    # Costruzione dataset
    X_list, y_z, y_xy = [], [], []
    for fp in files:
        try:
            df = pd.read_csv(fp, sep=";")
            if df.empty: continue
            row = df.iloc[0]
            X_list.append([float(row.get(gw, nan_fill)) for gw in gateways_order])
            y_z.append(int(round(float(row.get("z")))))
            y_xy.append([float(row.get("x")), float(row.get("y"))])
        except: continue

    X, Y_z, Y_xy = np.array(X_list), np.array(y_z), np.array(y_xy)

    # Step 1: Classificatore Piano (Z)
    log_fn(f"TRAIN: Fitting Piano Classifier (K={k_val})")
    floor_clf = KNeighborsClassifier(n_neighbors=k_val, weights=weights, metric=metric).fit(X, Y_z)
    
    # Step 2: Regressori X,Y per ogni piano trovato
    models_xy = {}
    for z in np.unique(Y_z):
        idx = np.where(Y_z == z)[0]
        log_fn(f"TRAIN: Fitting XY Regressor piano {z} ({len(idx)} campioni)")
        models_xy[int(z)] = KNeighborsRegressor(
            n_neighbors=min(k_val, len(idx)), 
            weights=weights, 
            metric=metric
        ).fit(X[idx], Y_xy[idx])

    # Salvataggio
    model_data = {
        "floor_clf": floor_clf,
        "xy_by_floor": models_xy,
        "gateways_order": gateways_order,
        "nan_fill": nan_fill,
        "created_at": datetime.now().isoformat()
    }
    
    os.makedirs(os.path.dirname(model_path), exist_ok=True)
    joblib.dump(model_data, model_path)
    log_fn(f"✅ TRAIN SUCCESS: Modello salvato in {model_path}")