MEDfl 2.0.10.dev9__tar.gz → 2.0.10.dev10__tar.gz

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/MEDfl/rw/client.py

@@ -13,6 +13,7 @@ from sklearn.model_selection import train_test_split
 from MEDfl.rw.model import Net  # votre définition de modèle
 from MEDfl.LearningManager.shap import LocalPyTorchSHAPExplainer
 from MEDfl.rw.holdout import run_local_holdout
+from MEDfl.rw.identity import client_name
 import socket
 import platform
 import psutil
@@ -426,7 +427,7 @@ class FlowerClient(fl.client.NumPyClient):
 
         # Convert NumPy arrays to lists before JSON serialization.
         serializable_result = {
-            "client_id": socket.gethostname(),
+            "client_id": client_name(),
             "feature_names": feature_names,
             "available_samples": int(len(loader.dataset)),
             "sample_count": explained_samples,
@@ -502,7 +503,7 @@ class FlowerClient(fl.client.NumPyClient):
             auc = float("nan")
         acc = accuracy_score(y_true, binary_preds)
 
-        hostname = socket.gethostname()
+        hostname = client_name()
         os_type = platform.system()
         metrics = {
             "hostname": hostname,
@@ -608,7 +609,7 @@ class FlowerClient(fl.client.NumPyClient):
                     features=self.effective_features,
                     target=self.effective_target,
                     predict=self._predict_probabilities,
-                    client_id=socket.gethostname(),
+                    client_id=client_name(),
                     drop_missing_features=True,
                     validation=self._validation_predictions,
                 )
@@ -636,7 +637,7 @@ class FlowerClient(fl.client.NumPyClient):
         return np.concatenate(y_parts), np.concatenate(prob_parts)
 
     def get_properties(self, config):
-        hostname = socket.gethostname()
+        hostname = client_name()
         os_type = platform.system()
 
         if self._initialized:

medfl-2.0.10.dev10/MEDfl/rw/identity.py

@@ -0,0 +1,12 @@
+"""
+Name a real-world client reports to the server (properties, metrics, SHAP and holdout results).
+
+The MEDfl agent ID (MEDFL_AGENT_ID, inherited from the agent that starts the client) when it is set,
+so several agents on the same machine are told apart. Else the machine hostname, as before.
+"""
+import os
+import socket
+
+
+def client_name() -> str:
+    return (os.getenv("MEDFL_AGENT_ID") or "").strip() or socket.gethostname()

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/MEDfl/rw/xgboost/client.py

@@ -31,6 +31,7 @@ try:
 except ImportError:
     GPUtil = None
 
+from MEDfl.rw.identity import client_name
 from MEDfl.rw.xgboost.data import XGBoostDataModule
 from MEDfl.rw.xgboost.metrics import evaluate_predictions
 from MEDfl.rw.xgboost.serialization import (
@@ -374,7 +375,7 @@ class XGBoostFlowerClient(fl.client.Client):
             )
 
         return {
-            "client_id": socket.gethostname(),
+            "client_id": client_name(),
             "feature_names": feature_names,
             "available_samples": available_samples,
             "sample_count": explained_samples,
@@ -457,7 +458,7 @@ class XGBoostFlowerClient(fl.client.Client):
         metrics = {
             "backend": "xgboost",
             "task": self.task,
-            "hostname": socket.gethostname(),
+            "hostname": client_name(),
             "os_type": platform.system(),
             "num_train": int(self.metadata.get("num_train", 0)),
             "features": ",".join(self.metadata.get("features", [])),
@@ -574,7 +575,7 @@ class XGBoostFlowerClient(fl.client.Client):
         if bool(config.get("holdout_enabled", False)):
             if self.task != "binary":
                 metrics["holdout_status"] = HOLDOUT_UNSUPPORTED
-                metrics["holdout_client_id"] = socket.gethostname()
+                metrics["holdout_client_id"] = client_name()
                 metrics["holdout_error"] = (
                     "The holdout test supports binary classification "
                     f"only (task is '{self.task}')"
@@ -587,7 +588,7 @@ class XGBoostFlowerClient(fl.client.Client):
                         features=self.data_module.effective_features,
                         target=self.data_module.effective_target,
                         predict=self._predict_probabilities,
-                        client_id=socket.gethostname(),
+                        client_id=client_name(),
                         drop_missing_features=False,
                         validation=self._validation_predictions,
                     )
@@ -622,7 +623,7 @@ class XGBoostFlowerClient(fl.client.Client):
         Return machine and dataset properties.
         """
 
-        hostname = socket.gethostname()
+        hostname = client_name()
         os_type = platform.system()
 
         df = self.data_module.df

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: medfl
-Version: 2.0.10.dev9
+Version: 2.0.10.dev10
 Summary: Python Open-source package for simulating federated learning and differential privacy
 Home-page: https://github.com/MEDomics-UdeS/MEDfl
 Author: MEDomics consortium

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/medfl.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: medfl
-Version: 2.0.10.dev9
+Version: 2.0.10.dev10
 Summary: Python Open-source package for simulating federated learning and differential privacy
 Home-page: https://github.com/MEDomics-UdeS/MEDfl
 Author: MEDomics consortium

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/medfl.egg-info/SOURCES.txt

@@ -38,6 +38,7 @@ MEDfl/NetManager/node.py
 MEDfl/rw/__init__.py
 MEDfl/rw/client.py
 MEDfl/rw/holdout.py
+MEDfl/rw/identity.py
 MEDfl/rw/model.py
 MEDfl/rw/server.py
 MEDfl/rw/strategy.py

{medfl-2.0.10.dev9 → medfl-2.0.10.dev10}/setup.py

@@ -14,7 +14,7 @@ with open('requirements.txt') as f:
 
 setup(
     name="medfl",
-    version="2.0.10.dev9",
+    version="2.0.10.dev10",
     author="MEDomics consortium",
     author_email="medomics.info@gmail.com",
     description="Python Open-source package for simulating federated learning and differential privacy",