Train a Classifier in Your Browser, Start to Finish

What you will build

A model that classifies the rows of a table, trained on your own machine with nothing installed. Any small labelled table works; the classic Iris flowers dataset is a good first choice.

Before you start

• A current version of Chrome or Edge
• A CSV file with one column that holds the label you want to predict

Step 1: Create a project

Open the Studio and choose New Project. A project keeps your data, preparation steps, model and results together, either in the browser or in a folder on your disk.

Step 2: Load the data

Open the Data Hub and add your CSV. It is parsed in the background and shown as a table, with a profile of every column: its type, how many values are missing, and how they are distributed. Nothing is uploaded.

Step 3: Prepare it

With the dataset open, choose PrepFlow from the Data Hub's tools in the sidebar. Build a short flow: drop duplicate rows, fill in missing values, scale the numeric columns, then split the rows into a training set and a test set. Preview the data as it moves through the flow to check each step did what you meant.

Step 4: Design the model

Open the Model Builder. Drag an input, two Dense layers and an output onto the canvas and connect them. Shapes are inferred as you connect, so a mismatch shows up on the canvas straight away, along with the parameter count.

Step 5: Train

Press Train. Training runs on your GPU through WebGPU, or WebGL where WebGPU is unavailable. Loss and accuracy update as each epoch finishes.

Step 6: Try it, then export

Open Deployment, type feature values into the playground and run inference. When the predictions look right, export the model.

Where to go next

• Generate the code for your model and keep editing it by hand: Visual Model Builder
• Make the whole sequence repeatable: Pipelines
• Chart the results: Dashboard