DAY 4 · TRAIN + TEST
Train, Test + Improve a Model
A model learns patterns from examples. It does not understand the object.
Topic · objective · success
Topic: Training and testing an image model
Objective: Train two classes, test new examples, and improve the model using evidence.
Success: Two fair classes, three new tests, one evidence-based improvement, and one human-check rule.
Open Teachable MachineDownload Mission Log · Word · Make a Google copyDownload Mission Log · PDF
Open AIR Lesson 3: Smart Machines in a new tab.
Collect → train → test → improve
- Choose two objects that are easy to distinguish.
- Collect varied training examples for both classes.
- Train the image model.
- Test three new examples that were not used for training.
- Record prediction, confidence, and whether it was correct.
- Change the training set based on the evidence.
- Name one situation where a human should decide.
Privacy + fairness
Use classroom objects, not faces or personal information. Keep the background and number of examples reasonably fair across both classes.
Confidence is not certainty. A high-confidence prediction can still be wrong.
Turn in · 10 points
Upload Page 3 or submit a clear photo of your test table and improvement.
Stem: “The model struggled when ___. We improved it by ___ because the test showed ___.”