Initial commit — Pseudonymisation de PDF v5

- GUI v5 : vue unique épurée (tkinter), 2 étapes visuelles
- Core ONNX : anonymisation regex + NER optionnel
- Extraction globale des noms depuis champs structurés
  (Patient, Rédigé par, MME/Madame, DR)
- Génération simultanée PDF Image + PDF Anonymisé (structure préservée)
- Build Windows via Nuitka (script batch + GitHub Actions CI)
- install.sh pour setup/run Linux

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

eds_pseudo_manager.py

@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+EDS-Pseudo Manager — Interface compatible NerModelManager pour le modèle AP-HP eds-pseudo.
+--------------------------------------------------------------------------------------------
+Utilise edsnlp pour charger le pipeline eds-pseudo (F1=0.97 sur données cliniques AP-HP).
+Mapping des 13 labels EDS-Pseudo vers les clés PLACEHOLDERS du core d'anonymisation.
+
+Dépendance : pip install 'edsnlp[ml]>=0.12.0'
+"""
+from __future__ import annotations
+from pathlib import Path
+from typing import Any, Dict, List, Optional
+
+try:
+    import edsnlp
+    _EDSNLP_AVAILABLE = True
+except ImportError:
+    edsnlp = None  # type: ignore
+    _EDSNLP_AVAILABLE = False
+
+# Mapping labels EDS-Pseudo → clés PLACEHOLDERS (anonymizer_core)
+EDS_LABEL_MAP: Dict[str, str] = {
+    "NOM": "NOM",
+    "PRENOM": "NOM",
+    "MAIL": "EMAIL",
+    "TEL": "TEL",
+    "SECU": "NIR",
+    "ADRESSE": "ADRESSE",
+    "ZIP": "CODE_POSTAL",
+    "VILLE": "VILLE",
+    "HOPITAL": "ETAB",
+    "DATE": "DATE",
+    "DATE_NAISSANCE": "DATE_NAISSANCE",
+    "IPP": "IPP",
+    "NDA": "NDA",
+}
+
+# Catalogue affiché dans la GUI
+EDS_MODELS_CATALOG: Dict[str, str] = {
+    "EDS-Pseudo AP-HP (edsnlp)": "AP-HP/eds-pseudo-public",
+}
+
+
+class EdsPseudoManager:
+    """Gestionnaire pour le modèle EDS-Pseudo (edsnlp). Même interface que NerModelManager."""
+
+    def __init__(self, cache_dir: Optional[Path] = None):
+        self.cache_dir = Path(cache_dir) if cache_dir else None
+        self.model_id: Optional[str] = None
+        self._nlp = None
+        self._loaded = False
+
+    def is_loaded(self) -> bool:
+        return self._loaded and self._nlp is not None
+
+    def load(self, model_id_or_path: str = "AP-HP/eds-pseudo-public") -> None:
+        if not _EDSNLP_AVAILABLE:
+            raise RuntimeError("edsnlp non disponible. Installez : pip install 'edsnlp[ml]>=0.12.0'")
+        self.unload()
+        self.model_id = model_id_or_path
+        path = Path(model_id_or_path)
+        if path.is_dir():
+            # Chargement local (modèle fine-tuné)
+            self._nlp = edsnlp.load(path)
+        else:
+            # Chargement depuis HuggingFace Hub
+            self._nlp = edsnlp.load(model_id_or_path)
+        self._loaded = True
+
+    def unload(self) -> None:
+        self._nlp = None
+        self._loaded = False
+        self.model_id = None
+
+    def models_catalog(self) -> Dict[str, str]:
+        return dict(EDS_MODELS_CATALOG)
+
+    def infer_paragraphs(
+        self,
+        paragraphs: List[str],
+        thresholds: Optional[Any] = None,
+        max_length: int = 384,
+        stride: int = 128,
+    ) -> List[List[Dict[str, Any]]]:
+        """Pour chaque paragraphe, retourne une liste d'entités détectées.
+
+        Chaque entité a les clés : entity_group, word, start, end, score, eds_mapped_key.
+        """
+        if not self.is_loaded():
+            return [[] for _ in paragraphs]
+
+        out: List[List[Dict[str, Any]]] = []
+        for para in paragraphs:
+            if not para.strip():
+                out.append([])
+                continue
+            doc = self._nlp(para)
+            ents: List[Dict[str, Any]] = []
+            for ent in doc.ents:
+                label = ent.label_.upper()
+                mapped = EDS_LABEL_MAP.get(label, None)
+                if mapped is None:
+                    continue
+                ents.append({
+                    "entity_group": label,
+                    "word": ent.text,
+                    "start": ent.start_char,
+                    "end": ent.end_char,
+                    "score": 1.0,  # edsnlp ne fournit pas de score de confiance
+                    "eds_mapped_key": mapped,
+                })
+            out.append(ents)
+        return out