← Back to course
AI Explorers 3-5 / Week 06 / Wednesday
3/6
Week 06 · Training Data

Wednesday

Explorer Lab: Train a paper model
// Learning from labeled examples
⏱ about 15 min

Wednesday: Explorer Lab: Train a Paper Model

Rocket and Raven make 16 photo cards from the training data table.

"Now we train the model," Rocket says. "It learns by counting!"

"What do you notice when we tally the features?" Raven asks.

Nova hovers over the cards. "Count carefully," she hints. "The tally is what your model learns."

Your mission

You will be the learning model.

First, you make the training data. Then you tally it to find patterns.

Last, you use the patterns to sort four new cards.

  • 20 index cards or paper squares
  • A pencil and crayons
  • Sticky notes or small paper scraps for labels
  • The training data table from Tuesday
Safety first
Use only paper, pencils and crayons.
Carry pencils with the point down.
Clean up your cards when you finish.
  1. Make 16 cards that match Tuesday's training data table. Write or draw the features.
  2. Put the right label on each card.
  3. Make a tally chart: features down the side, labels across the top.
  4. Make one tally mark for each feature on each card.
  5. Circle any feature that shows up for only one label. That is a pattern!
  6. Ask a grown-up to make 4 new cards with no labels.
  7. Use your patterns to predict a label for each new card.
Featuretoolplantrock
a handle
straight edges
shiny
leaves
lumpy
For a grown-up
For the 4 new cards, use these features.
Card A: shiny metal, no handle, straight edges. Card B: green, leaves.
Card C: shiny and wet, gray, lumpy. Card D: gray, lumpy.
PREDICT THE NEW CARDS
  • Read the question.
  • Tap your answer.
Card A: shiny metal, no handle, straight edges. Your prediction?
Card B: green, leaves. Your prediction?
Card C: shiny and wet, gray, lumpy. Your prediction?
Card D: gray, lumpy. Your prediction?
New cardMy predictionGrown-up saysRight? (yes or no)
Atool
Bplant
Crock
Drock
How many tally marks go in the tool column for a handle?
How many tally marks go in the rock column for shiny?
How many cards did you make for the training data?
What we learnedTrue or false?
The tally chart showed which features go with which label.?
Card A, the ruler card, was predicted as tool.?
The model needed labels to learn the patterns.?
The model could predict new cards without any training data.?
WHY THIS EXERCISEThe model learned patterns from labeled training data, then used them on new cards.
How many of the 4 new cards did you predict right? Type a number.
WHY THIS EXERCISEAny score teaches you something about your model.

You trained a model with paper and pencils! Tomorrow we find out why more data helps.

← Tuesday