Dev

README.md

@@ -1,5 +1,2 @@
-<p style="text-align: center;">
-  <img src="logo_OCR.jpg" alt="Logo Academy" width="100">
-</p>
-#Anticipez_les_besoins_en_consommation_de_batiments
-Projet ML pour prédire la consommation d’énergie des bâtiments. Parcours complet : exploration et nettoyage des données, préparation des variables, entraînement et comparaison de modèles, sélection du meilleur, sauvegarde, API de prédiction, tests et exemples.
+
+

bentofile.yaml

@@ -0,0 +1,14 @@
+service: "service:EnergyAPI"
+
+include:
+  - "service.py"
+
+python:
+  packages:
+    - numpy
+    - pandas
+    - scikit-learn==1.5.2     # même version que le training
+    - bentoml==1.4.25
+
+models:
+  - "energy_model:latest"

check_keys.py

@@ -0,0 +1,8 @@
+import json, bentoml
+ref = bentoml.models.get("energy_model:latest")
+expected = list(ref.custom_objects["feature_names"])
+with open("request.json","r",encoding="utf-8") as f:
+    got = list(json.load(f).keys())
+print("== same length? ", len(expected)==len(got), len(expected))
+print("Missing:", sorted(set(expected)-set(got)))
+print("Extra  :", sorted(set(got)-set(expected)))

make.ps1

@@ -0,0 +1,51 @@
+param([Parameter(Position=0)][string]$Task="help")
+
+function Train { python .\train_and_save.py; bentoml models list }
+function Build { bentoml build; bentoml containerize energy_api:latest -t energy_api:latest }
+function Run {
+  docker ps -q --filter "publish=3000" | ForEach-Object { docker rm -f $_ } | Out-Null
+  docker rm -f energy_api 2>$null | Out-Null
+  docker run -d --name energy_api -p 3000:3000 energy_api:latest
+  docker ps --filter "name=energy_api"
+}
+
+function Logs  { docker logs -f energy_api }
+function Stop  { docker rm -f energy_api }
+function Inspect {
+  $payload = Get-Content .\request.json -Raw | ConvertFrom-Json
+  $json = (@{ payload = $payload } | ConvertTo-Json -Depth 10 -Compress)
+  $enc  = New-Object System.Text.UTF8Encoding($false)   # UTF-8 SANS BOM
+  $bytes = $enc.GetBytes($json)
+  Invoke-RestMethod -Uri "http://localhost:3000/inspect" -Method Post -ContentType "application/json; charset=utf-8" -Body $bytes
+}
+function Predict {
+  $payload = Get-Content .\request.json -Raw | ConvertFrom-Json
+  $json = (@{ payload = $payload } | ConvertTo-Json -Depth 10 -Compress)
+  $enc  = New-Object System.Text.UTF8Encoding($false)   # UTF-8 SANS BOM
+  $bytes = $enc.GetBytes($json)
+  Invoke-RestMethod -Uri "http://localhost:3000/predict" -Method Post -ContentType "application/json; charset=utf-8" -Body $bytes
+}function Help {
+  @"
+Usage: .\make.ps1 <task>
+
+Tasks:
+  train       — entraîne + sauvegarde le modèle
+  build       — bento build + docker image
+  run         — lance le conteneur en arrière-plan (port 3000)
+  logs        — suit les logs du conteneur
+  stop        — stoppe & supprime le conteneur
+  inspect     — envoie request.json à /inspect (enveloppe auto sans BOM)
+  predict     — envoie request.json à /predict (enveloppe auto sans BOM)
+"@
+}
+
+switch ($Task) {
+  "train"   { Train }
+  "build"   { Build }
+  "run"     { Run }
+  "logs"    { Logs }
+  "stop"    { Stop }
+  "inspect" { Inspect }
+  "predict" { Predict }
+  default   { Help }
+}

make_payload.py

@@ -0,0 +1,53 @@
+import json, bentoml
+
+ref = bentoml.models.get("energy_model:latest")
+features = list(ref.custom_objects.get("feature_names", []))
+
+payload = {}
+for f in features:
+    lf = f.lower()
+
+    # one-hot PropertyCategory_*
+    if f.startswith("PropertyCategory_"):
+        payload[f] = 1 if f == "PropertyCategory_Office" else 0
+        continue
+
+    # ratios
+    if "ratio" in lf or "fraction" in lf or "share" in lf:
+        payload[f] = 0.7
+        continue
+
+    # surfaces / GFA
+    if any(k in lf for k in ["gfa","area","surface","size","sqft","sf"]):
+        if "parking" in lf: payload[f] = 30000
+        elif "second" in lf: payload[f] = 40000
+        elif "largest" in lf: payload[f] = 180000
+        elif "building(s" in lf or "building_s" in lf: payload[f] = 220000
+        else: payload[f] = 250000
+        continue
+
+    # étages
+    if "floor" in lf or "floors" in lf or "story" in lf:
+        payload[f] = 5
+        continue
+
+    # âge / années
+    if "âge" in f.lower() or "age" in lf or "year" in lf or "years" in lf:
+        payload[f] = 40
+        continue
+
+    # nombre de bâtiments
+    if "numberofbuildings" in lf or "nbuildings" in lf or "buildings" in lf:
+        payload[f] = 1
+        continue
+
+    # défaut
+    payload[f] = 100.0
+
+# ASSURE la présence explicite de la clé accentuée
+payload.setdefault("Surface moyenne par étage", 250000)
+
+with open("request.json","w",encoding="utf-8") as f:
+    json.dump(payload, f, ensure_ascii=False, indent=2)
+
+print("OK -> request.json (", len(payload), "keys )")

raccourci_cmd_cleansheet.txt

@@ -0,0 +1,107 @@
+1		venv311 : .\.venv311\Scripts\Activate.ps1
+
+2		lancer le serveur dans le venv311 : python -m bentoml serve service:svc --reload
+
+3		ouvrir le swagger : http://127.0.0.1:3000
+
+4		Regénère un request.json propre (0/1 pour les dummies)
+
+		Exécute dans la racine du projet (venv activé) :
+
+@'
+import json, bentoml, unicodedata as ud
+
+ref = bentoml.models.get("energy_model:latest")
+features = list(ref.custom_objects.get("feature_names", []))
+
+def default_value(name: str):
+    n = name.lower()
+    # dummies
+    if name.startswith("PropertyCategory_"):
+        return 1 if name == "PropertyCategory_Office" else 0
+    # ratios
+    if "ratio" in n or "fraction" in n or "share" in n:
+        return 0.7
+    # surfaces / gfa
+    if any(k in n for k in ["gfa","area","surface","size","sqft","sf"]):
+        if "parking" in n: return 30000
+        if "second" in n: return 40000
+        if "largest" in n: return 180000
+        if "building(s" in n or "building_s" in n: return 220000
+        return 250000
+    # étages
+    if "floor" in n or "floors" in n or "story" in n:
+        return 5
+    # âge / années
+    if "age" in n or "année" in n or "an " in n or "year" in n or "years" in n:
+        return 40
+    # nombre de bâtiments
+    if "numberofbuildings" in n or "nbuildings" in n or "buildings" in n:
+        return 1
+    # défaut
+    return 100.0
+
+# construit le payload EN UTILISANT EXACTEMENT les clés du modèle
+payload = { name: default_value(name) for name in features }
+
+with open("request.json","w",encoding="utf-8") as f:
+    json.dump(payload, f, ensure_ascii=False, indent=2)
+
+print("OK -> request.json aligné, nb clés:", len(payload))
+'@ | Out-File -Encoding utf8 rebuild_payload.py
+
+python .\rebuild_payload.py
+type .\request.json
+
+
+DOCKER
+
+
+
+ Tag                          Size       Model Size  Creation Time
+ energy_api:2l7mkre5qo7n3l7m  42.29 KiB  0.00 B      2025-09-30 00:29:56
+
+
+
+Successfully built Bento(tag="energy_api:72idth45rc3hhl7m").
+
+
+
+
+
+
+
+2) Lance l’inspect & la prédiction
+.\make.ps1 inspect
+.\make.ps1 predict
+
+Option rapide (sans éditer de fichier)
+
+Tu peux envoyer un bâtiment “à la volée” :
+
+$payload = @{
+  BuildingAge = 40; NumberofBuildings = 1; NumberofFloors = 5
+  PropertyGFATotal = 250000; PropertyGFAParking = 30000; "PropertyGFABuilding(s)" = 220000
+  LargestPropertyUseTypeGFA = 180000; SecondLargestPropertyUseTypeGFA = 40000
+  PropertyCategory = "Office"
+}
+$json = (@{ payload = $payload } | ConvertTo-Json -Depth 10 -Compress)
+$enc  = New-Object System.Text.UTF8Encoding($false)
+$bytes = $enc.GetBytes($json)
+Invoke-RestMethod -Uri "http://localhost:3000/predict" -Method Post -ContentType "application/json; charset=utf-8" -Body $bytes
+
+
+
+
+1) Pré-prod locale (checklist rapide)
+
+Rebuild si tu changes service.py ou réentraînes :
+
+.\make.ps1 train      # si tu réentraînes
+.\make.ps1 build
+.\make.ps1 run
+.\make.ps1 health
+.\make.ps1 predict
+
+
+Garde un request.json “type” (celui qui marche) dans le repo.

rebuild_payload.py

@@ -0,0 +1,39 @@
+import json, bentoml, unicodedata as ud
+
+ref = bentoml.models.get("energy_model:latest")
+features = list(ref.custom_objects.get("feature_names", []))
+
+def default_value(name: str):
+    n = name.lower()
+    # dummies
+    if name.startswith("PropertyCategory_"):
+        return 1 if name == "PropertyCategory_Office" else 0
+    # ratios
+    if "ratio" in n or "fraction" in n or "share" in n:
+        return 0.7
+    # surfaces / gfa
+    if any(k in n for k in ["gfa","area","surface","size","sqft","sf"]):
+        if "parking" in n: return 30000
+        if "second" in n: return 40000
+        if "largest" in n: return 180000
+        if "building(s" in n or "building_s" in n: return 220000
+        return 250000
+    # étages
+    if "floor" in n or "floors" in n or "story" in n:
+        return 5
+    # âge / années
+    if "age" in n or "année" in n or "an " in n or "year" in n or "years" in n:
+        return 40
+    # nombre de bâtiments
+    if "numberofbuildings" in n or "nbuildings" in n or "buildings" in n:
+        return 1
+    # défaut
+    return 100.0
+
+# construit le payload EN UTILISANT EXACTEMENT les clés du modèle
+payload = { name: default_value(name) for name in features }
+
+with open("request.json","w",encoding="utf-8") as f:
+    json.dump(payload, f, ensure_ascii=False, indent=2)
+
+print("OK -> request.json aligné, nb clés:", len(payload))

request.json

@@ -0,0 +1,13 @@
+{
+  "BuildingAge": 40,
+  "Surface moyenne par étage": 44000,
+  "Ratio Largest Use / Total": 0.81818,
+  "NumberofBuildings": 1,
+  "NumberofFloors": 5,
+  "PropertyGFATotal": 250000,
+  "PropertyGFAParking": 30000,
+  "PropertyGFABuilding(s)": 220000,
+  "LargestPropertyUseTypeGFA": 180000,
+  "SecondLargestPropertyUseTypeGFA": 40000,
+  "PropertyCategory": "Office"
+}

request_wrapped.json

@@ -0,0 +1,20 @@
+{"payload": {
+  "BuildingAge": 40,
+  "Surface moyenne par étage": 250000,
+  "Ratio Largest Use / Total": 0.7,
+  "NumberofBuildings": 1,
+  "NumberofFloors": 5,
+  "PropertyGFATotal": 250000,
+  "PropertyGFAParking": 30000,
+  "PropertyGFABuilding(s)": 220000,
+  "LargestPropertyUseTypeGFA": 180000,
+  "SecondLargestPropertyUseTypeGFA": 40000,
+  "PropertyCategory_Healthcare": 0,
+  "PropertyCategory_Hospitality": 0,
+  "PropertyCategory_Industrial & Storage": 0,
+  "PropertyCategory_Mixed Use": 0,
+  "PropertyCategory_Office": 1,
+  "PropertyCategory_Other": 0,
+  "PropertyCategory_Public / Institutional": 0,
+  "PropertyCategory_Retail & Food": 0
+} }

sanity_check.py

@@ -0,0 +1,41 @@
+# sanity_check.py
+import json, pandas as pd, numpy as np, bentoml
+
+CSV_PATH = "ML_modele_iqr_cat.csv"
+TARGET = "SiteEnergyUse(kBtu)"
+FEATURES = [
+    "BuildingAge","Surface moyenne par étage","Ratio Largest Use / Total",
+    "NumberofBuildings","NumberofFloors","PropertyGFATotal","PropertyGFAParking",
+    "PropertyGFABuilding(s)","LargestPropertyUseTypeGFA","SecondLargestPropertyUseTypeGFA",
+    "PropertyCategory",
+]
+
+# 1) repères
+df = pd.read_csv(CSV_PATH)
+q = df[TARGET].quantile([0.05,0.25,0.5,0.75,0.95]).to_dict()
+print("[Target quantiles kBtu]", {int(k*100): float(v) for k,v in q.items()})
+
+# 2) modèle
+ref = bentoml.models.get("energy_model:latest")
+model = bentoml.sklearn.load_model(ref)
+md = ref.info.metadata or {}
+use_log = bool(md.get("use_log_target", False))
+
+# 3) payload
+with open("request.json","r",encoding="utf-8") as f:
+    p = json.load(f)
+
+# reconstruire la catégorie si one-hot
+if "PropertyCategory" not in p:
+    cat_cols = [c for c in p if c.startswith("PropertyCategory_")]
+    if not cat_cols:
+        raise SystemExit("Aucune colonne PropertyCategory_*.")
+    best = max(cat_cols, key=lambda c: float(p.get(c,0) or 0))
+    p["PropertyCategory"] = best.split("PropertyCategory_",1)[1]
+
+X = pd.DataFrame([p], columns=FEATURES)
+y = model.predict(X)
+if use_log: y = np.expm1(y)
+
+print("[Request used]", X.iloc[0].to_dict())
+print("[Prediction kBtu]", float(np.ravel(y)[0]))