🧼 Nettoyage des chemins, chargement via Hub

app.py

@@ -19,11 +19,11 @@ from src.utils.alert_email import send_alert_email
 from transformers import pipeline as hf_pipeline, AutoModelForSequenceClassification, AutoTokenizer
 import plotly.express as px
 
-# === Chargement du modèle depuis le Model Registry ===
-# model = mlflow.transformers.load_model("models:/emotions_classifier/Production")
-# model_path = "models/electra_model"
-model = AutoModelForSequenceClassification.from_pretrained("arnaud66170/electra-p9v2.1")
-tokenizer = AutoTokenizer.from_pretrained("arnaud66170/electra-p9v2.1")
+# === Chargement du modèle hébergé sur Hugging Face Hub ===
+MODEL_REPO = "arnaud66170/electra-p9v2.1"
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_REPO)
+tokenizer = AutoTokenizer.from_pretrained(MODEL_REPO)
+emotion_pipeline = hf_pipeline("text-classification", model=model, tokenizer=tokenizer, top_k=None, function_to_apply="sigmoid")
 
 # === Globals ===
 HISTORY_LIMIT = 5
@@ -32,8 +32,8 @@ ALERT_WINDOW_MINUTES = 5
 ALERT_COOLDOWN_MINUTES = 10
 FEEDBACK_ALERT_THRESHOLD = 3
 alert_history = []
+
 FEEDBACK_CSV = os.path.abspath("feedback_log_emotions.csv")
-# EXPORT_CSV_PATH = os.path.abspath("outputs/emotion_stats_2000.csv")
 EXPORT_XLSX_PATH = os.path.abspath("outputs/emotion_summary.xlsx")
 suspect_text = ""
 
@@ -87,12 +87,6 @@ def contains_intense_words(text):
     words = re.findall(r"\w+", text.lower())
     return any(w in INTENSE_WORDS for w in words)
 
-model_path = Path("notebooks/models/emotions/model").resolve().as_posix()
-tokenizer_path = Path("notebooks/models/emotions/tokenizer").resolve().as_posix()
-local_model = AutoModelForSequenceClassification.from_pretrained(model_path)
-tokenizer = AutoTokenizer.from_pretrained(tokenizer_path)
-emotion_pipeline = hf_pipeline("text-classification", model=local_model, tokenizer=tokenizer, top_k=None, function_to_apply="sigmoid")
-
 def predict_emotions(text):
     global suspect_text
     if not text.strip():
@@ -107,7 +101,7 @@ def predict_emotions(text):
     alert_banner = ""
     if contains_intense_words(text):
         suspect_text = text
-        alert_banner = "🟥 <strong style='color:red;'>⚠️ Ce tweet contient un mot intense. Consultez l’onglet \\\"Cas suspects\\\" pour audit.</strong>"
+        alert_banner = "🟥 <strong style='color:red;'>⚠️ Ce tweet contient un mot intense. Consultez l’onglet \"Cas suspects\" pour audit.</strong>"
     html_output = f"<h2 style='text-align:center;'>🧠 Résultat : {result}</h2>" + (f"<div style='margin-top:10px'>{alert_banner}</div>" if alert_banner else "")
     return html_output, update_pie_chart(), update_history(), text if alert_banner else ""
 
@@ -180,7 +174,6 @@ def update_feedback_stats():
     except:
         return "Erreur lecture stats."
 
-
 def generate_emotion_summary():
     if not history:
         return None
@@ -234,9 +227,4 @@ with gr.Blocks() as demo:
     gen_button.click(fn=generate_emotion_summary, outputs=download_btn)
 
 if __name__ == "__main__":
-    demo.launch()
-
-# # Lancement de l'interface
-# 
-# python app.py
-
+    demo.launch()