Evaluate and Improve

Although the teachable machine does not allow you to changes its underlying algorithm, you can tweak the parameters of the learning process to improve  model performance.  

---

Practice Goals:

For this practice session, focus on interpreting the training results to understand if the model has trained too much (overfit) or too little (underfit), and make sure it is trained just right (balanced fit) by adjusting the number of epochs.  After tuning the learning process, validate how the model performs on previously unseen data.  Put on your testers hat for a moment and try to select interesting test images.  

---

Hands-On Activities:

• Validate the Training Results:
  Click [Under the Hood]
  View [Accuracy/Loss Per Epoch]
  
• Retrain Model for Balanced Fit:
  Modify [Epochs]  
• Evaluate Model Predictions:
  Toggle [Input Switch] On
  Set [Input Dropdown] to [File]
  Click [Choose Image]
  Select Image from [val] folder
  View Classification [Output]
  

---

How-To Guides:

Validating the Training Results and Improving Model Performance

Leveraging the Image Classification Model  to Make Predictions