Skip to main content

Setup

Course progressStage 0 of 10
~45 min
Your workspace

Keep your Colab notebook tab open all session. Open in a new tab — don’t use the buttons in this page to leave the course.

Train your AIGoogle ColabOpen in a new tab ↗No-code warm-upMachine Learning for KidsOpen in a new tab ↗Open the starter notebookOpen in ColabOpen in a new tab ↗
Build

your own copy of the starter notebook in Google Colab, running on a GPU

Learn

what Colab is and how a notebook runs code one cell at a time

Ship

a notebook that prints Hello and is ready for Stage 1

Teacher demo

Before students touch their laptops, show the room on the projector:

  1. Click the Open in Colab button. Show that it opens the starter notebook.
  2. Do File → Save a copy in Drive so everyone has their own copy to edit.
  3. Set Runtime → Change runtime type → GPU, then run the print cell with Shift + Enter.
  4. Make the point: a notebook is a stack of cells we run top to bottom, and the work lives in Google Drive — closing the tab doesn't lose it.

The big idea​

Google Colab is a free coding notebook that runs in your browser. You write Python in little boxes called cells and run them one at a time. Colab does the heavy lifting on Google's computers — including a GPU, a chip that trains AI models fast — so your laptop doesn't have to.

This course is a first machine-learning path through one project: an image classifier. You will use Python tools to inspect image data, prepare it, build a model, train it, grade it honestly, study mistakes, improve the experiment, and explain the evidence.

A notebook is not one long program. It's a stack of cells, and you run them top to bottom. A cell can use a variable made in an earlier cell, which is powerful — but it also means that if the notebook disconnects (which happens), those variables disappear and you have to run the earlier cells again.

How the Python ML workflow connects
  1. 1
    Photos / CIFAR-10labeled image examplesStage 1
  2. 2
    Notebook variablesx_train, y_train, class_namesSetup-2
  3. 3
    Prepared datanormalized pixels and fair pilesStage 3
  4. 4
    Keras modelCNN layers and summaryStage 4
  5. 5
    Training historyepochs, loss, accuracyStage 5
  6. 6
    Test evidencesealed score and mistakesStages 6-7
  7. 7
    Improved modelaugmentation comparisonStage 8
  8. 8
    Inferenceuploaded image to top-3 guessesStage 9
  9. 9
    Demo evidencetable, confidence, limitationStage 10

Setup creates the workspace that holds the whole workflow. Every later stage adds one more notebook result: data, prepared arrays, model, training history, test evidence, predictions, and final demo.

┌─────────────────────────────┐
│  cell 1   load the data     │  ← run this first
├─────────────────────────────┤
│  cell 2   build the model   │  ← then this
├─────────────────────────────┤
│  cell 3   train the model   │  ← then this
└─────────────────────────────┘
   run top → bottom, every time
New words
Colab
a free notebook in your browser where you write and run Python
cell
one box in the notebook that holds code you can run
runtime
the computer Colab gives you to run your code
GPU
a chip that trains AI models much faster than a normal one

Build it​

Step 1 — Open the starter notebook​

Click the Open in Colab button at the top of this page. A notebook called image-lab-starter opens. It already has a section for every stage of the week.

Step 2 — Save your own copy​

Right now you're looking at a read-only copy. Make it yours: click File → Save a copy in Drive.

A new tab opens titled Copy of image-lab-starter. This is the one you edit all week. Rename it to something you'll recognize, like YourName_PythonML, by clicking the title at the top.

From now on, the Open in Colab button is only for getting a fresh copy. Your saved copy lives in your Google Drive — find it again at drive.google.com.

Step 3 — Turn on the GPU​

Click Runtime → Change runtime type. Choose GPU, then Save.

Training without a GPU still works — it's just slower. Turning it on now means Stage 4 finishes in a couple of minutes instead of ten.

Step 4 — Run your first cell​

Find the first empty code cell and type:

print("Hello! My Python machine learning notebook is ready.")

Run it by clicking the ▶ button on the left of the cell, or by pressing Shift + Enter. The first run takes a few seconds while Colab wakes up the runtime. You should see your message appear right below the cell.

Test your setup​

  • You have your own copy of the notebook (the title says Copy of… or your new name, not just image-lab-starter).
  • The runtime type is set to GPU.
  • Your print cell ran and showed your message below it.
  • You can find your saved notebook again in Google Drive.
  • Workflow check. Point to the workflow map and explain why Colab is the workspace for the whole course.
  • Design check. Tell a neighbor the difference between a text cell, a code cell, and an output. If you can explain all three, you understand the tool.

If it breaks​

  • The Open in Colab button shows a 404 or "not found." The starter notebook may not be published yet — ask your coach for the shared notebook link instead, then do File → Save a copy in Drive.
  • It asks me to sign in. Colab needs a Google account. Use the one your camp gave you, or your own.
  • My cell shows a red error. Read the last line of the error first — it usually names the problem. For Setup, the most common cause is a typo in print.
  • "Cannot connect to GPU backend." Free GPUs are sometimes busy. Switch Runtime type back to CPU for now; everything still works, just slower. Try the GPU again later.
  • My notebook says "disconnected." Colab disconnects if you leave it idle. Click Reconnect, then Runtime → Run all to re-run your cells.
Coach notes

The single biggest time sink here is students editing the read-only starter instead of their own copy — their work vanishes when they reload. Walk the room after Step 2 and confirm every title reads Copy of….

If your camp pre-creates Google accounts, hand out the credentials before opening Colab; signing in eats 5–10 minutes otherwise. GPU backends can be unavailable at peak times — it's fine to run the whole course on CPU at this dataset size if you have to; only Stage 4 feels the difference.