Update docs/Training-ML-Agents.md

Co-authored-by: Vincent-Pierre BERGES <vincentpierre@unity3d.com>

docs/Training-ML-Agents.md

 [Unity Inference Engine](Getting-Started.md#running-a-pre-trained-model).
 
 Additionally, if the network architecture changes, you may still load an existing model,
-and ML-Agents will only load the parts of the model that haven't changed. For instance,
+but ML-Agents will only load the parts of the model it can load and ignore all others. For instance,
 if you add a new reward signal, the existing model will load but the new reward signal
 will be initialized from scratch. If you have a model with a visual encoder (CNN) but
 change the `hidden_units`, the CNN will be loaded but the body of the network will be