Day 3: How Machines Learn

Recap

  • Learned about Pandas
  • Created charts with Matplotlib
  • Investigated data on Ulaanbaatar’s air quality

But, we did all the work ourselves!

How can we use AI to detect the patterns in the data?

Introducing scikit-learn

What is scikit-learn?

  • A Python library for machine learning
  • Built on top of Pandas — works with DataFrames you already know
  • Provides ready-to-use models for:
    • Regression — predicting a number (price, temperature, score)
    • Classification — predicting a category (spam/not spam, yes/no)
    • Clustering — finding hidden groups in data

Using scikit-learn

  • Every sklearn model follows the same three steps:
  1. Prepare your features (X) and labels (y)
  2. Fit — model.fit(X, y) — the model learns from your data
  3. Predict — model.predict(X_new) — apply what it learned to new data

Demo

Open and run the “Using Scikit-learn” Notebook (sklearn.ipynb)

Hands-on

Open and run the “Using Scikit-learn” Notebook (sklearn.ipynb)

Reflection

You created your first AI model!

Let’s use some real world data!

Dzud

  • For many families in Mongolia, livestock are the primary source of food, income, and cultural identity.
  • Periodically, a combination of summer drought and extreme winter cold kills millions of animals in a single season.
  • This event is called a dzud.
  • In 2001 alone, over seven million animals died. Entire herding families were left with nothing.

Video

https://www.youtube.com/watch?v=GZiXZU9asNQ

Can we use AI to predict an upcoming dzud?

Demo

Open and run the “Finding Disasters” Notebook (finding_disasters.ipynb)

Hands-on

Open and run the “Finding Disasters” Notebook (finding_disasters.ipynb)

Reflection

What did you discover? Did anything surprise you?

Finding The Cause

Finding The Cause

  • In the previous notebook you found the disasters.
    • 2001 and 2010 were catastrophic years
    • Southern and western aimags tend to suffer more than others
  • This tells use the when and where but not the why

Demo and Walk Through

Open and run the “Finding The Cause” Notebook (finding_cause.ipynb)

Let’s Train Our Model!

Let’s Train Our Model!

  • Winter temperature is the strongest predictor of livestock mortality
  • A prior summer drought increases mortality
  • Can a machine learn that relationship?
  • And then predict mortality for future years?
  • This is called linear regression

What is Linear Regression?

  • It draws the best-fit straight line through your data points
  • Once the line is drawn, you can predict new values
  • “Linear” just means the relationship is a straight line

What is Linear Regression?

Training and Test Datasets

  • We split our data into two groups:
    • A training set to teach the model, and a test set to check how well it learned
  • The test set is like an exam at the end to see if the model is working
  • A common split is 80% training / 20% testing

Demo

Open and run the “Linear Regression” Notebook (regression.ipynb)

Hands-On

Open and run the “Linear Regression” Notebook (regression.ipynb)

Reflection

Congratulations! You trained your first AI model!

Tomorrow

  • Use the same approach for hardware
  • Explore micro:bit boards
  • Use AI to detect waving, putting your hand up, and other commands!