Running telemetry completely in the background

src/tabpfn_common_utils/telemetry/core/events.py

@@ -245,10 +245,10 @@ class BaseTelemetryEvent:
     """
 
     # Python version that the SDK is running on
-    python_version: str = field(default_factory=_get_py_version, init=False)
+    python_version: Optional[str] = field(default=None, init=False)
 
     # TabPFN version that the SDK is running on
-    tabpfn_version: str = field(default_factory=_get_sdk_version, init=False)
+    tabpfn_version: Optional[str] = field(default=None, init=False)
 
     # Timestamp of the event
     timestamp: datetime = field(default_factory=_utc_now, init=False)
@@ -257,18 +257,32 @@ class BaseTelemetryEvent:
     extension: Optional[str] = field(default=None, init=False)
 
     # Runtime environment of the platform
-    runtime_kernel: Optional[str] = field(
-        default_factory=_get_runtime_kernel, init=False
-    )
+    runtime_kernel: Optional[str] = field(default=None, init=False)
 
     # Runtime environment of the platform
-    runtime_environment: Optional[str] = field(
-        default_factory=_get_runtime_environment, init=False
-    )
+    runtime_environment: Optional[str] = field(default=None, init=False)
 
     # Operating system of the platform
     platform_os: str = field(default_factory=_get_platform_os, init=False)
 
+    def enrich(self):
+        """Enrich the event with additional properties.
+
+        We enrich the events with additional properties so we do not
+        have to call the default factories at initialization time.
+        """
+        if self.runtime_kernel is None:
+            self.runtime_kernel = _get_runtime_kernel()
+
+        if self.runtime_environment is None:
+            self.runtime_environment = _get_runtime_environment()
+
+        if self.python_version is None:
+            self.python_version = _get_py_version()
+
+        if self.tabpfn_version is None:
+            self.tabpfn_version = _get_sdk_version()
+
     @property
     def name(self) -> str:
         raise NotImplementedError
@@ -378,22 +392,22 @@ class ModelCallEvent(BaseTelemetryEvent):
     task: Literal["classification", "regression"]
 
     # Version of the PyTorch
-    torch_version: str = field(default_factory=_get_torch_version, init=False)
+    torch_version: Optional[str] = field(default=None, init=False)
 
     # Version of the scikit-learn
-    sklearn_version: str = field(default_factory=_get_sklearn_version, init=False)
+    sklearn_version: Optional[str] = field(default=None, init=False)
 
     # Version of the NumPy
-    numpy_version: str = field(default_factory=_get_numpy_version, init=False)
+    numpy_version: Optional[str] = field(default=None, init=False)
 
     # Version of the Pandas
-    pandas_version: str = field(default_factory=_get_pandas_version, init=False)
+    pandas_version: Optional[str] = field(default=None, init=False)
 
     # Version of the AutoGluon
-    autogluon_version: str = field(default_factory=_get_autogluon_version, init=False)
+    autogluon_version: Optional[str] = field(default=None, init=False)
 
     # Type of GPU if available
-    gpu_type: Optional[str] = field(default_factory=_get_gpu_type, init=False)
+    gpu_type: Optional[str] = field(default=None, init=False)
 
     # Version of the model
     model_version: Optional[str] = field(default=None, init=False)
@@ -410,6 +424,28 @@ class ModelCallEvent(BaseTelemetryEvent):
     # Duration of the model call in milliseconds
     duration_ms: int = -1
 
+    def enrich(self):
+        """Enrich the event with additional properties."""
+        super().enrich()
+
+        if self.gpu_type is None:
+            self.gpu_type = _get_gpu_type()
+
+        if self.torch_version is None:
+            self.torch_version = _get_torch_version()
+
+        if self.sklearn_version is None:
+            self.sklearn_version = _get_sklearn_version()
+
+        if self.numpy_version is None:
+            self.numpy_version = _get_numpy_version()
+
+        if self.pandas_version is None:
+            self.pandas_version = _get_pandas_version()
+
+        if self.autogluon_version is None:
+            self.autogluon_version = _get_autogluon_version()
+
 
 @dataclass
 class FitEvent(ModelCallEvent):