--- title: "Train a recycling sorter" subtitle: "micro:bit and the computer · A model drives a build · Lesson 4" --- # Train a recycling sorter **Name:** {{field-24}} **Date:** {{field-15}} **Class:** {{field-15}} > **Key question:** Recycling centres use cameras and trained models to sort paper from plastic on a fast conveyor belt. Before a centre trusts one, someone has to test it on items it has never seen. :::teacher-note **In one sentence:** Students train a Paper/Plastic image model in Teachable Machine, test it on photos it has never seen, and fix it by changing the training data. **Students should be able to say:** I can train an image model and test it on photos it has never seen. I can work out accuracy, overall and for each group of test photos. I can find what a model's mistakes share and fix it by changing its training data. **Width:** the unmarked parts are the floor every student does (predict, run, investigate and make). Typical adds the modify practice; wider adds the real-world contexts. They add contexts, not difficulty. Students work in the online tutorial alongside this sheet. ::: ## Predict :::keep-together **Your model gets your own items right almost every time on the webcam. The board tests it on other paper and plastic photographed on a messy bench. Its score on the messy bench will be…** ☐ Much lower than on plain backgrounds ☐ Exactly 50%: it guesses ☐ About the same: it has learned what paper is ☐ Higher: more objects help it My prediction, and why: {{lines-3}} ::: :::teacher-note **What happens:** Much lower. The model learned from photos on one plain background, so the background is part of what it learned. On a messy bench it is often wrong. ::: ## Run Run the program in the tutorial. Was your prediction right? What was different? {{lines-2}} ## Read the results :::keep-together ### Accuracy **A model gets 23 of 32 test photos right. What is its accuracy, as a percentage (to the nearest whole number)?** Answer: {{field-15}} ::: :::teacher-note **Answer:** 72 Accuracy is correct ÷ total: 23 ÷ 32 = 0.72, or 72%. ::: :::keep-together ### What the failures share **This model scores: plain 8/8, messy 4/8, low light 5/8, not paper or plastic 0/8. Its training photos were all on one plain desk. Why does it fail on the messy bench?** ☐ It needs more training epochs ☐ It cannot tell the item from the background it was trained on ☐ Messy photos are harder for every model, so nothing can be done ☐ It has forgotten what paper is ::: :::teacher-note **Answer:** It cannot tell the item from the background it was trained on Every training photo had the same plain desk, so the desk is part of what the model learned. On a cluttered bench it is guessing. ::: :::keep-together ### Your turn **Another model scores well everywhere except low light. Which fix is most likely to work?** ☐ Rename the classes ☐ Add more photos on the same bright desk ☐ Add training photos taken in low light ☐ Train for longer ::: :::teacher-note **Answer:** Add training photos taken in low light Look for what the failures share (low light), then add training photos that have it. Change the data, not the settings. ::: :::keep-together ### Sam's model **Sam's model scores 50% on the messy bench. Sam adds 100 more photos of the same items on the same desk. The messy score doesn't change. What went wrong?** ☐ The new photos were like the old ones, so they added no variety ☐ The board's test is unfair ☐ Sam needed to delete the old photos first ☐ 100 photos is too many ::: :::teacher-note **Answer:** The new photos were like the old ones, so they added no variety More of the same data teaches the model nothing new. Sam needs photos on different backgrounds. ::: :::typical ## Fix the data **Backgrounds.** Retrain your model with photos of the items on **at least three different backgrounds** (or upload the **varied photos**). Export, paste the new link and **Load**. **Unbalanced classes.** **Neither.** Look at the **not paper or plastic** group: the model gives a can or a jar a label anyway, often with high confidence. A model always picks one of its classes. Add a third class called **Neither** with photos of other things (cans, glass, fruit), retrain and test again. Work in the tutorial. Plan first: what will you change, and how will you test it? {{box-4}} ::: ## Diagnose :::keep-together ### Midday photos **A model was trained on photos taken at midday by a window. It scores 100% plain, 88% messy, 25% low light. What is the problem, and the fix?** ☐ Too few classes: add a Neither class ☐ No dim photos in training: add photos taken in low light ☐ Too many photos: delete some ☐ The test is unfair: ignore low light ::: :::teacher-note **Answer:** No dim photos in training: add photos taken in low light ::: :::keep-together ### A confident can **A model scores 90% on paper and plastic but labels every can as Plastic with 95% confidence. What is the problem, and the fix?** ☐ High confidence means it is right: nothing to fix ☐ Cans are too shiny for any model ☐ It has no class for other things: add a Neither class ☐ It needs more plastic photos ::: :::teacher-note **Answer:** It has no class for other things: add a Neither class ::: :::keep-together ### Lopsided classes **Paper has 80 photos and Plastic has 15. The model calls most plastic items Paper. What is the problem, and the fix?** ☐ The classes are unbalanced: give each class about the same number of photos ☐ Plastic is harder to see: add a filter ☐ Retrain with the same photos ☐ It needs a Neither class ::: :::teacher-note **Answer:** The classes are unbalanced: give each class about the same number of photos ::: ## Make a sorter you would trust Use what you found: varied backgrounds and light, balanced classes, and a **Neither** class. Then ask a classmate to bring three items your model has never seen. Does it still work? Work in the tutorial. Plan first: what will you change, and how will you test it? {{box-4}} :::wider ## Evaluate Write one or two sentences: which items or conditions your sorter would still get wrong, and what a real recycling centre would need to test before trusting it. **Revision summary** - A model learns only from the examples you give it: it works on inputs like its training data and fails on others. - **Accuracy** = correct ÷ total. Test on photos the model has never seen, and look at each **group**, not just the overall score. - Find what the failures share, then **change the data**: more variety, balanced classes, a class for "none of these". - A model always picks one of its classes, and it can be confident and wrong. ::: ## Exit ticket :::keep-together **A model scores 95% on its own desk and 50% in the canteen. What do you do?** {{lines-3}} ::: :::teacher-note **Answer:** Find what the canteen photos have that the training photos don't (background, light), and add training photos like that. :::