{ "cells": [ { "cell_type": "markdown", "id": "7afa1a06", "metadata": { "id": "7afa1a06" }, "source": [ "**Projet UE8 : tester de nouvelles couches de données (hors rendu)**\n", "\n", "Notebook séparé pour ne pas toucher au rendu. Il relit les ventes nettoyées (`data/alpes_terrain.csv`, produites par le notebook principal) et compare le Random Forest réglé (200 arbres, 4 variables par embranchement) en validation croisée (10 plis), sans puis avec chaque nouvelle couche.\n", "\n", "Couches testées, une par une :\n", "1. Prix des ventes voisines : testé hors Colab le 09/10, gain négligeable (43,6 -> 42,9 €/m²).\n", "2. Distance aux remontées mécaniques (OpenStreetMap).\n", "3. Zonage du PLU (Géoportail de l'urbanisme, API Carto IGN).\n", "4. Parcelle cadastrale, bâti et ensoleillement : voir `Projet_UE8_couches_test_2.ipynb`." ] }, { "cell_type": "markdown", "id": "d91d232d", "metadata": { "id": "d91d232d" }, "source": [ "# Setup" ] }, { "cell_type": "code", "execution_count": 1, "id": "fe7b4b16", "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "fe7b4b16", "executionInfo": { "status": "ok", "timestamp": 1791536978616, "user_tz": -120, "elapsed": 16997, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "eb72c6d2-079b-40cd-cf28-630430e0c42a" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Mounted at /content/drive\n", "Dossier : /content/drive/MyDrive/Colab Notebooks/Projet Habitat 05\n" ] } ], "source": [ "from google.colab import drive\n", "drive.mount('/content/drive')\n", "\n", "import os, time, json, requests\n", "import numpy as np, pandas as pd\n", "import geopandas as gpd\n", "from shapely.geometry import LineString\n", "from sklearn.ensemble import RandomForestRegressor\n", "from sklearn.model_selection import KFold\n", "\n", "PROJECT_ROOT_DIR = \"/content/drive/MyDrive/Colab Notebooks/Projet Habitat 05\"\n", "os.chdir(PROJECT_ROOT_DIR)\n", "assert os.path.exists(os.path.join(\"data\", \"alpes_terrain.csv\")), \"⚠️ data/alpes_terrain.csv introuvable : relancer le notebook principal\"\n", "print(\"Dossier :\", os.getcwd())" ] }, { "cell_type": "markdown", "id": "feedcf29", "metadata": { "id": "feedcf29" }, "source": [ "## Ventes et variables de base\n", "\n", "Mêmes variables que le modèle principal. `nb_ventes_commune` est recalculé à l'intérieur de chaque pli, sur l'entraînement seulement (pas de fuite)." ] }, { "cell_type": "code", "execution_count": 2, "id": "301b41bb", "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 98 }, "id": "301b41bb", "executionInfo": { "status": "ok", "timestamp": 1791536999415, "user_tz": -120, "elapsed": 20786, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "5c7981d2-ab41-43ec-b734-d686313a5a32" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "4137 ventes géolocalisées\n" ] }, { "output_type": "execute_result", "data": { "text/plain": [ " RMSE MAE Erreur relative médiane Ventes à moins de 15 % \\\n", "Référence 43.3 25.9 14% 52% \n", "\n", " RMSE stations \n", "Référence 99.4 " ], "text/html": [ "\n", "
\n", "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
RMSEMAEErreur relative médianeVentes à moins de 15 %RMSE stations
Référence43.325.914%52%99.4
\n", "
\n", "
\n", "\n", "
\n", " \n", "\n", " \n", "\n", " \n", "
\n", "\n", "
\n", "
\n" ], "application/vnd.google.colaboratory.intrinsic+json": { "type": "dataframe", "summary": "{\n \"name\": \"pd\",\n \"rows\": 1,\n \"fields\": [\n {\n \"column\": \"RMSE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 43.3,\n \"max\": 43.3,\n \"num_unique_values\": 1,\n \"samples\": [\n 43.3\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"MAE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 25.9,\n \"max\": 25.9,\n \"num_unique_values\": 1,\n \"samples\": [\n 25.9\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Erreur relative m\\u00e9diane\",\n \"properties\": {\n \"dtype\": \"string\",\n \"num_unique_values\": 1,\n \"samples\": [\n \"14%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Ventes \\u00e0 moins de 15 %\",\n \"properties\": {\n \"dtype\": \"string\",\n \"num_unique_values\": 1,\n \"samples\": [\n \"52%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"RMSE stations\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 99.4,\n \"max\": 99.4,\n \"num_unique_values\": 1,\n \"samples\": [\n 99.4\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n }\n ]\n}" } }, "metadata": {}, "execution_count": 2 } ], "source": [ "df = pd.read_csv(os.path.join(\"data\", \"alpes_terrain.csv\"),\n", " dtype={\"code_departement\": str, \"code_commune\": str}).dropna(subset=[\"longitude\"]).reset_index(drop=True)\n", "print(len(df), \"ventes géolocalisées\")\n", "\n", "def km(lon1, lat1, lon2, lat2): # distance à vol d'oiseau (haversine)\n", " lon1, lat1, lon2, lat2 = map(np.radians, [lon1, lat1, lon2, lat2])\n", " a = np.sin((lat2 - lat1) / 2) ** 2 + np.cos(lat1) * np.cos(lat2) * np.sin((lon2 - lon1) / 2) ** 2\n", " return 6371 * 2 * np.arcsin(np.sqrt(a))\n", "\n", "PREFECTURES = {\"04\": (6.2357, 44.0925), \"05\": (6.0794, 44.5594)}\n", "POLES = [(6.0616, 44.5797), (5.7896, 43.8293), (6.2495, 44.0954), (6.6536, 44.8995)] # Gap, Manosque, Digne, Briançon\n", "df[\"dist_prefecture_km\"] = [km(lo, la, *PREFECTURES[d]) for lo, la, d in zip(df[\"longitude\"], df[\"latitude\"], df[\"code_departement\"])]\n", "df[\"dist_pole_km\"] = np.min([km(df[\"longitude\"], df[\"latitude\"], *p) for p in POLES], axis=0)\n", "df[\"log_population\"] = np.log(df[\"population_commune\"])\n", "df[\"log_surface\"] = np.log(df[\"surface\"])\n", "\n", "BASE = [\"annee\", \"longitude\", \"latitude\", \"surface\", \"part_AB\", \"population_commune\", \"altitude\", \"pente_pct\",\n", " \"orientation_sud\", \"dist_cours_eau_m\", \"dist_prefecture_km\", \"dist_pole_km\", \"log_population\", \"log_surface\"]\n", "CATEGORIES = [\"code_departement\", \"type_mixite\", \"nom_epci\"]\n", "y = df[\"prix_m2\"].values\n", "STATIONS = [\"Montgenèvre\", \"Briançon\", \"Saint-Chaffrey\", \"La Salle-les-Alpes\", \"Le Monêtier-les-Bains\",\n", " \"Vars\", \"Risoul\", \"Les Orres\", \"Orcières\", \"Dévoluy\"]\n", "en_station = df[\"nom_commune\"].isin(STATIONS).values\n", "\n", "def evaluer(numeriques=[], categories=[]):\n", " \"\"\"Validation croisée 10 plis du Random Forest réglé, avec les colonnes ajoutées.\"\"\"\n", " X = pd.concat([df[BASE + numeriques], pd.get_dummies(df[CATEGORIES + categories]).astype(int)], axis=1)\n", " pred = np.zeros(len(y))\n", " for tr, te in KFold(n_splits=10, shuffle=True, random_state=42).split(X):\n", " Xtr, Xte = X.iloc[tr].copy(), X.iloc[te].copy()\n", " nb = df.iloc[tr].groupby(\"code_commune\").size()\n", " Xtr[\"nb_ventes_commune\"] = df[\"code_commune\"].iloc[tr].map(nb).values\n", " Xte[\"nb_ventes_commune\"] = df[\"code_commune\"].iloc[te].map(nb).fillna(0).values\n", " rf = RandomForestRegressor(n_estimators=200, max_features=4, random_state=42, n_jobs=-1).fit(Xtr, y[tr])\n", " pred[te] = rf.predict(Xte)\n", " err, rel = pred - y, np.abs(pred - y) / y\n", " return {\"RMSE\": round(np.sqrt((err ** 2).mean()), 1), \"MAE\": round(np.abs(err).mean(), 1),\n", " \"Erreur relative médiane\": f\"{np.median(rel):.0%}\", \"Ventes à moins de 15 %\": f\"{(rel < 0.15).mean():.0%}\",\n", " \"RMSE stations\": round(np.sqrt((err[en_station] ** 2).mean()), 1)}\n", "\n", "resultats = {\"Référence\": evaluer()}\n", "pd.DataFrame(resultats).T" ] }, { "cell_type": "markdown", "id": "5b010c3c", "metadata": { "id": "5b010c3c" }, "source": [ "## Couche 2 : distance aux remontées mécaniques (OpenStreetMap)\n", "\n", "Toutes les remontées (télésièges, télécabines, téléskis...) du 04 et du 05, en une seule requête à l'API Overpass, mises en cache dans `data/remontees_osm.geojson`.\n", "Deux variables : distance à la remontée la plus proche (en mètres) et nombre de remontées à moins de 5 km (taille du domaine skiable)." ] }, { "cell_type": "code", "execution_count": 4, "id": "63b95fea", "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "63b95fea", "executionInfo": { "status": "ok", "timestamp": 1791537114603, "user_tz": -120, "elapsed": 2808, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "b0778c75-2291-4c68-e16c-2dd75b468e4c" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "Réponse de https://overpass-api.de/api/interpreter\n", "825 remontées mécaniques\n", "type\n", "platter 281\n", "chair_lift 224\n", "magic_carpet 117\n", "drag_lift 106\n", "gondola 43\n", "rope_tow 17\n", "mixed_lift 16\n", "cable_car 10\n", "t-bar 7\n", "j-bar 4\n", "Name: count, dtype: int64\n" ] } ], "source": [ "REMONTEES_PATH = os.path.join(\"data\", \"remontees_osm.geojson\")\n", "OVERPASS = [\"https://overpass-api.de/api/interpreter\",\n", " \"https://maps.mail.ru/osm/tools/overpass/api/interpreter\",\n", " \"https://overpass.kumi.systems/api/interpreter\"]\n", "ENTETES = {\"User-Agent\": \"Projet-UE8-MBA-terrains-Alpes/1.0 (Google Colab)\", \"Accept\": \"application/json\"}\n", "TYPES = \"chair_lift|gondola|cable_car|mixed_lift|drag_lift|t-bar|j-bar|platter|rope_tow|magic_carpet\"\n", "BBOX = \"43.66,5.49,45.13,7.08\" # sud, ouest, nord, est : 04 + 05\n", "\n", "if os.path.exists(REMONTEES_PATH):\n", " remontees = gpd.read_file(REMONTEES_PATH)\n", "else:\n", " requete = f'[out:json][timeout:180];way[\"aerialway\"~\"^({TYPES})$\"]({BBOX});out geom;'\n", " data = None\n", " for essai in range(2): # 2 tours sur les 3 serveurs\n", " for url in OVERPASS:\n", " try:\n", " r = requests.post(url, data={\"data\": requete}, headers=ENTETES, timeout=200)\n", " r.raise_for_status()\n", " data = r.json()\n", " print(\"Réponse de\", url)\n", " break\n", " except Exception as e:\n", " print(f\"⚠️ Échec sur {url} : {e}\")\n", " if data is not None:\n", " break\n", " print(\"Nouvel essai dans 60 s...\")\n", " time.sleep(60)\n", " if data is None:\n", " raise RuntimeError(\"⚠️ Aucun serveur Overpass n'a répondu : réessayer dans quelques minutes\")\n", " lignes = [{\"type\": w[\"tags\"].get(\"aerialway\"), \"nom\": w[\"tags\"].get(\"name\"),\n", " \"geometry\": LineString([(p[\"lon\"], p[\"lat\"]) for p in w[\"geometry\"]])}\n", " for w in data[\"elements\"] if len(w.get(\"geometry\", [])) >= 2]\n", " if not lignes:\n", " raise RuntimeError(\"⚠️ Réponse reçue mais aucune remontée trouvée : vérifier la zone (BBOX)\")\n", " remontees = gpd.GeoDataFrame(lignes, crs=\"EPSG:4326\")\n", " remontees.to_file(REMONTEES_PATH, driver=\"GeoJSON\")\n", "\n", "print(len(remontees), \"remontées mécaniques\")\n", "print(remontees[\"type\"].value_counts())" ] }, { "cell_type": "code", "execution_count": 5, "id": "c44ffe1d", "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 442 }, "id": "c44ffe1d", "executionInfo": { "status": "ok", "timestamp": 1791537151736, "user_tz": -120, "elapsed": 20775, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "69f111f7-20f0-48f1-c4b8-a4b0fdecea3d" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ " dist_remontee_m nb_remontees_5km\n", "count 4137.0 4137.0\n", "mean 13741.0 4.0\n", "std 11705.0 9.0\n", "min 25.0 0.0\n", "25% 4213.0 0.0\n", "50% 8890.0 0.0\n", "75% 21175.0 5.0\n", "max 50736.0 59.0\n", "\n", "Prix médian selon la distance à la remontée la plus proche :\n", " ventes prix_median\n", "dist_remontee_m \n", "(0.0, 500.0] 254 136.0\n", "(500.0, 1000.0] 172 154.0\n", "(1000.0, 2000.0] 182 109.0\n", "(2000.0, 5000.0] 572 111.0\n", "(5000.0, 10000.0] 1046 116.0\n", "(10000.0, 1000000.0] 1911 108.0\n" ] }, { "output_type": "execute_result", "data": { "text/plain": [ " RMSE MAE Erreur relative médiane \\\n", "Référence 43.3 25.9 14% \n", "+ remontées mécaniques 42.9 25.5 14% \n", "\n", " Ventes à moins de 15 % RMSE stations \n", "Référence 52% 99.4 \n", "+ remontées mécaniques 53% 99.5 " ], "text/html": [ "\n", "
\n", "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
RMSEMAEErreur relative médianeVentes à moins de 15 %RMSE stations
Référence43.325.914%52%99.4
+ remontées mécaniques42.925.514%53%99.5
\n", "
\n", "
\n", "\n", "
\n", " \n", "\n", " \n", "\n", " \n", "
\n", "\n", "
\n", "
\n" ], "application/vnd.google.colaboratory.intrinsic+json": { "type": "dataframe", "summary": "{\n \"name\": \"pd\",\n \"rows\": 2,\n \"fields\": [\n {\n \"column\": \"RMSE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 42.9,\n \"max\": 43.3,\n \"num_unique_values\": 2,\n \"samples\": [\n 42.9,\n 43.3\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"MAE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 25.5,\n \"max\": 25.9,\n \"num_unique_values\": 2,\n \"samples\": [\n 25.5,\n 25.9\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Erreur relative m\\u00e9diane\",\n \"properties\": {\n \"dtype\": \"string\",\n \"num_unique_values\": 1,\n \"samples\": [\n \"14%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Ventes \\u00e0 moins de 15 %\",\n \"properties\": {\n \"dtype\": \"string\",\n \"num_unique_values\": 2,\n \"samples\": [\n \"53%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"RMSE stations\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 99.4,\n \"max\": 99.5,\n \"num_unique_values\": 2,\n \"samples\": [\n 99.5\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n }\n ]\n}" } }, "metadata": {}, "execution_count": 5 } ], "source": [ "# Distances en mètres : on passe en Lambert 93 (EPSG:2154), le système métrique officiel en France\n", "points = gpd.GeoDataFrame(df[[\"longitude\", \"latitude\"]], geometry=gpd.points_from_xy(df[\"longitude\"], df[\"latitude\"]),\n", " crs=\"EPSG:4326\").to_crs(\"EPSG:2154\")\n", "lignes_l93 = remontees.to_crs(\"EPSG:2154\")\n", "\n", "proche = gpd.sjoin_nearest(points, lignes_l93[[\"geometry\"]], distance_col=\"d\")\n", "df[\"dist_remontee_m\"] = proche.groupby(level=0)[\"d\"].min()\n", "tampon = points.copy(); tampon[\"geometry\"] = points.buffer(5000)\n", "df[\"nb_remontees_5km\"] = gpd.sjoin(tampon, lignes_l93[[\"geometry\"]], predicate=\"intersects\").groupby(level=0).size()\n", "df[\"nb_remontees_5km\"] = df[\"nb_remontees_5km\"].fillna(0)\n", "\n", "print(df[[\"dist_remontee_m\", \"nb_remontees_5km\"]].describe().round())\n", "print(\"\\nPrix médian selon la distance à la remontée la plus proche :\")\n", "print(df.groupby(pd.cut(df[\"dist_remontee_m\"], [0, 500, 1000, 2000, 5000, 10000, 1e6]), observed=True)[\"prix_m2\"]\n", " .agg(ventes=\"count\", prix_median=\"median\").round())\n", "\n", "resultats[\"+ remontées mécaniques\"] = evaluer([\"dist_remontee_m\", \"nb_remontees_5km\"])\n", "pd.DataFrame(resultats).T" ] }, { "cell_type": "markdown", "id": "d533fc2b", "metadata": { "id": "d533fc2b" }, "source": [ "## Couche 3 : zonage du PLU (Géoportail de l'urbanisme)\n", "\n", "Pour chaque vente, on demande à l'API Carto de l'IGN la zone d'urbanisme à cet endroit :\n", "- **PLU** : U (urbanisé), AU (à urbaniser), A (agricole), N (naturel) ;\n", "- **carte communale** : secteur constructible ou non ;\n", "- rien : la commune est au **règlement national d'urbanisme** (RNU).\n", "\n", "⚠️ C'est le zonage **actuel**, pas celui de la date de la vente. Environ 4 100 requêtes, mises en cache dans `data/plu_cache.csv` (compter 15 à 30 minutes au premier lancement ; si la cellule s'arrête, la relancer : elle reprend où elle en était)." ] }, { "cell_type": "code", "execution_count": 9, "id": "bff63b44", "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "bff63b44", "executionInfo": { "status": "ok", "timestamp": 1791539026792, "user_tz": -120, "elapsed": 685, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "d26746af-7e7f-4454-8529-b5aa4dd3639a" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ "1 points à interroger\n", "1/1 points, 0 échecs\n" ] } ], "source": [ "PLU_PATH = os.path.join(\"data\", \"plu_cache.csv\")\n", "APICARTO = \"https://apicarto.ign.fr/api/gpu\"\n", "\n", "def zone_en(lon, lat, session):\n", " geom = json.dumps({\"type\": \"Point\", \"coordinates\": [lon, lat]})\n", " r = session.get(f\"{APICARTO}/zone-urba\", params={\"geom\": geom}, timeout=30)\n", " r.raise_for_status()\n", " f = r.json().get(\"features\", [])\n", " if f:\n", " return \"PLU_\" + str(f[0][\"properties\"].get(\"typezone\", \"?\"))\n", " r = session.get(f\"{APICARTO}/secteur-cc\", params={\"geom\": geom}, timeout=30)\n", " r.raise_for_status()\n", " f = r.json().get(\"features\", [])\n", " if f:\n", " return \"CC_\" + str(f[0][\"properties\"].get(\"typesect\", \"?\"))\n", " return \"RNU\"\n", "\n", "cache = pd.read_csv(PLU_PATH) if os.path.exists(PLU_PATH) else pd.DataFrame(columns=[\"lon_r\", \"lat_r\", \"zone\"])\n", "df[\"lon_r\"], df[\"lat_r\"] = df[\"longitude\"].round(5), df[\"latitude\"].round(5)\n", "deja = set(zip(cache[\"lon_r\"], cache[\"lat_r\"]))\n", "a_faire = df[[\"lon_r\", \"lat_r\"]].drop_duplicates()\n", "a_faire = a_faire[[(lo, la) not in deja for lo, la in zip(a_faire[\"lon_r\"], a_faire[\"lat_r\"])]]\n", "print(len(a_faire), \"points à interroger\")\n", "\n", "nouveaux, echecs = [], 0\n", "with requests.Session() as s:\n", " for n, (lo, la) in enumerate(zip(a_faire[\"lon_r\"], a_faire[\"lat_r\"]), 1):\n", " try:\n", " nouveaux.append({\"lon_r\": lo, \"lat_r\": la, \"zone\": zone_en(lo, la, s)})\n", " except Exception as e:\n", " echecs += 1\n", " if echecs <= 5:\n", " print(f\"⚠️ Échec ({lo}, {la}) : {e}\")\n", " if n % 200 == 0 or n == len(a_faire): # sauvegarde régulière\n", " cache = pd.concat([cache, pd.DataFrame(nouveaux)], ignore_index=True); nouveaux = []\n", " cache.to_csv(PLU_PATH, index=False)\n", " print(f\"{n}/{len(a_faire)} points, {echecs} échecs\")\n", " time.sleep(0.05)\n", "\n", "if echecs:\n", " print(f\"⚠️ {echecs} points sans réponse : relancer la cellule pour les reprendre\")" ] }, { "cell_type": "code", "execution_count": 10, "id": "29138391", "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 331 }, "id": "29138391", "executionInfo": { "status": "ok", "timestamp": 1791539070593, "user_tz": -120, "elapsed": 40462, "user": { "displayName": "Nathalie Wirth", "userId": "04968417715143734623" } }, "outputId": "40ab76e2-3d32-4f59-cd94-cc7bef43ec38" }, "outputs": [ { "output_type": "stream", "name": "stdout", "text": [ " ventes prix_median\n", "zone_plu \n", "CC_01 98 57.0\n", "CC_03 18 30.0\n", "PLU_A 93 70.0\n", "PLU_AU 527 124.0\n", "PLU_N 63 62.0\n", "PLU_U 2814 121.0\n", "RNU 524 78.0\n" ] }, { "output_type": "execute_result", "data": { "text/plain": [ " RMSE MAE Erreur relative médiane \\\n", "Référence 43.3 25.9 14% \n", "+ remontées mécaniques 42.9 25.5 14% \n", "+ zonage PLU 43.3 25.9 14% \n", "+ remontées + PLU 43.3 25.7 14% \n", "\n", " Ventes à moins de 15 % RMSE stations \n", "Référence 52% 99.4 \n", "+ remontées mécaniques 53% 99.5 \n", "+ zonage PLU 53% 100.5 \n", "+ remontées + PLU 53% 101.0 " ], "text/html": [ "\n", "
\n", "
\n", "\n", "\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
RMSEMAEErreur relative médianeVentes à moins de 15 %RMSE stations
Référence43.325.914%52%99.4
+ remontées mécaniques42.925.514%53%99.5
+ zonage PLU43.325.914%53%100.5
+ remontées + PLU43.325.714%53%101.0
\n", "
\n", "
\n", "\n", "
\n", " \n", "\n", " \n", "\n", " \n", "
\n", "\n", "
\n", "
\n" ], "application/vnd.google.colaboratory.intrinsic+json": { "type": "dataframe", "summary": "{\n \"name\": \"pd\",\n \"rows\": 4,\n \"fields\": [\n {\n \"column\": \"RMSE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 42.9,\n \"max\": 43.3,\n \"num_unique_values\": 2,\n \"samples\": [\n 42.9,\n 43.3\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"MAE\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 25.5,\n \"max\": 25.9,\n \"num_unique_values\": 3,\n \"samples\": [\n 25.9,\n 25.5\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Erreur relative m\\u00e9diane\",\n \"properties\": {\n \"dtype\": \"category\",\n \"num_unique_values\": 1,\n \"samples\": [\n \"14%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"Ventes \\u00e0 moins de 15 %\",\n \"properties\": {\n \"dtype\": \"string\",\n \"num_unique_values\": 2,\n \"samples\": [\n \"53%\"\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n },\n {\n \"column\": \"RMSE stations\",\n \"properties\": {\n \"dtype\": \"date\",\n \"min\": 99.4,\n \"max\": 101.0,\n \"num_unique_values\": 4,\n \"samples\": [\n 99.5\n ],\n \"semantic_type\": \"\",\n \"description\": \"\"\n }\n }\n ]\n}" } }, "metadata": {}, "execution_count": 10 } ], "source": [ "df = df.drop(columns=[\"zone\"], errors=\"ignore\").merge(cache, on=[\"lon_r\", \"lat_r\"], how=\"left\")\n", "# Regroupement : U, AU (AUc, AUs...), A, N, carte communale constructible / non, RNU\n", "def famille(z):\n", " if pd.isna(z): return \"inconnu\"\n", " if z.startswith(\"PLU_AU\"): return \"PLU_AU\"\n", " if z.startswith(\"PLU_U\"): return \"PLU_U\"\n", " if z.startswith(\"PLU_A\"): return \"PLU_A\"\n", " if z.startswith(\"PLU_N\"): return \"PLU_N\"\n", " return z\n", "df[\"zone_plu\"] = df[\"zone\"].map(famille)\n", "print(df.groupby(\"zone_plu\")[\"prix_m2\"].agg(ventes=\"count\", prix_median=\"median\").round())\n", "\n", "resultats[\"+ zonage PLU\"] = evaluer(categories=[\"zone_plu\"])\n", "if \"dist_remontee_m\" in df.columns:\n", " resultats[\"+ remontées + PLU\"] = evaluer([\"dist_remontee_m\", \"nb_remontees_5km\"], [\"zone_plu\"])\n", "pd.DataFrame(resultats).T" ] }, { "cell_type": "markdown", "id": "c85521fc", "metadata": { "id": "c85521fc" }, "source": [ "## Bilan\n", "\n", "Référence : RMSE 43,3 €/m² en validation croisée (10 plis), 99,4 €/m² dans les communes de stations.\n", "\n", "| Couche | RMSE (€/m²) | RMSE stations | Lecture |\n", "|---|---|---|---|\n", "| Prix des ventes voisines (testé à part) | 42,9 au mieux | | Le Random Forest fait déjà cette moyenne locale avec les coordonnées |\n", "| Remontées mécaniques (OpenStreetMap) | 42,9 | 99,5 | Environ +30 % à moins d'1 km d'une remontée, mais déjà capté par la position |\n", "| Zonage du PLU (Géoportail de l'urbanisme) | 43,3 | 100,5 | Zones U et AU vers 121-124 €/m², A et N vers 62-70 €/m² ; 81 % des ventes sont en U ou AU |\n", "\n", "Aucune couche de localisation n'améliore le modèle : il connaît déjà l'emplacement. L'erreur restante vient de ce que DVF ne dit pas du terrain lui-même. Suite : couches « parcelle » dans `Projet_UE8_couches_test_2.ipynb`." ] } ], "metadata": { "colab": {}, "kernelspec": { "display_name": "Python 3", "name": "python3" }, "language_info": { "name": "python" } }, "nbformat": 4, "nbformat_minor": 5 }