Code state 1
Train
model.fit(measurements, labels)A model turns inputs into predictions
Ask for evidence that an AI works on new cases.
Diagram explanation
A learned model maps measurements to a species prediction.
- Flower measurements leads to Learned model
- Learned model leads to Species prediction
- Model
- The learned part of a system that maps inputs to predictions.
Examples give it something to learn from
| Sepal length | Sepal width | Petal length | Petal width | Known species |
|---|---|---|---|---|
| 5.1 | 3.5 | 1.4 | 0.2 | setosa |
- Inputs / features
- The four measurements.
- Label
- The known species we want the model to predict.
- Supervised learning
- Learning from examples with target answers.
Training changes it; using it makes a prediction
TRAIN · make
- Labeled examples → fit a model
- The model is changed.
INFER · use
- New measurements → prediction
- The fitted model is used.
Code state 2
Use
model.predict(new_measurements)Learn the pattern, not the quirks
Conceptual fitting patterns; shapes identify groups. Not measured Iris results.
CS 450 · original conceptual illustration · Original course schematic; all rights reserved
Check cases kept out of training
TRAINING FLOWERS
- Used to fit the model
- 112 examples
HELD-OUT FLOWERS
- Kept out while fitting
- 38 examples
Diagram explanation
The score summarizes predictions on the separate cases.
- Predict species leads to Compare with kept-back label
- Compare with kept-back label leads to Count correct / checked
- Accuracy
- Correct predictions divided by the number of cases checked.
Learning can also use rewards for actions
SUPERVISED LEARNING
- Examples with target answers
- Iris measurements + species
REINFORCEMENT LEARNING
- Actions produce reward feedback
- Game moves → game reward
Follow the same story in Iris
Diagram explanation
Separate cases provide a check on a trained model.
- Examples leads to Train
- Train leads to Model
- Model leads to Use + check
- Decision tree
- Questions about measurements lead to a species prediction.
- Depth
- How many levels of questions the tree can use.
- Try depth 1, then depth 3.
- Explain the new result.
Open the Iris notebook
Get a working notebook or join the paper/projected route.
- Open the Iris activity from the CS 450 course page.
- In Colab, choose Runtime → Run all.
- Find a code cell and the output underneath it.
- Notebook: blkfin.github.io/cs450/
Read the inputs, prediction, and check
Explain one example and one prediction.
- Find the four input measurements and the species label.
- Find the 38 flowers kept out of training.
- At depth 1, read the correct count and denominator.
- Choose one displayed case: does its prediction match its label?
- Depth 1: 25 of 38 correct · 65.8%
Allow more questions, then check again
Predict what could change before you rerun.
- In Step 6, change TREE_DEPTH from 1 to 3.
- Run that cell and every cell below it.
- Compare the new correct count on the same held-out flowers.
- Depth 3: 37 of 38 correct · 97.4%
Keep the claim within the evidence
Complete three lines.
- Training did __; inference did __.
- The model got __ of __ held-out flowers right.
- That does not establish __.
- Ask what it learned from, how it was checked, and whether the check fits your use.