model: pytorch: Updated to include is_trained flag

model/pytorch/dffml_model_pytorch/pytorch_accuracy_scorer.py

@@ -24,7 +24,7 @@ class PytorchAccuracyContext(AccuracyContext):
     async def score(
         self, mctx: ModelContext, sctx: Sources, *features: Features
     ):
-        if not mctx.parent.model_path.exists():
+        if not mctx.is_trained:
             raise ModelNotTrained("Train model before assessing for accuracy.")
 
         dataset, size = await mctx.dataset_generator(sctx)

model/pytorch/dffml_model_pytorch/pytorch_base.py

@@ -298,14 +298,15 @@ async def train(self, sources: Sources):
                 "Best Validation Accuracy: {:4f}".format(best_acc)
             )
             self.parent.model.load_state_dict(best_model_wts)
+        self.is_trained = True
 
     async def predict(
         self, sources: SourcesContext
     ) -> AsyncIterator[Tuple[Record, Any, float]]:
         """
         Uses trained data to make a prediction about the quality of a record.
         """
-        if not self.parent.model_path.exists():
+        if not self.is_trained:
             raise ModelNotTrained("Train model before prediction.")
 
         self.parent.model.eval()
@@ -399,6 +400,7 @@ async def __aenter__(self) -> "PyTorchModel":
         if self.model_path.exists():
             self.logger.info(f"Using saved model from {self.model_path}")
             self.model = torch.load(self.model_path)
+            self.is_trained = True
         else:
             self.model = self.createModel()
         return self