GitHub Copilot
Use an AI explanation to investigate a coding bug without skipping the reasoning.
Explore this workflow →Interactive lessons · Education-focused
TensorFlow Playground
Explore a small neural network visually using built-in two-dimensional datasets and adjustable training settings.
Open TensorFlow Playground ↗Official getting-started help ↗For: Teachers demonstrating neural networks and supervised learners ready to discuss graphs and model error. Documentation reviewed November 7, 2026.
Is it a good fit for your task?
Choose it when…
Teacher demonstrations of features, decision boundaries and training versus test performance.
Consider another approach when…
Training a classroom surveillance model or treating a toy experiment as a benchmark for real-world AI.
A practical first workflow
Observe how one model change affects a simple classification experiment.
Have this ready
Open the Playground on a display and prepare a table for settings, training loss and test loss.
- Choose a built-in classification dataset and select the x1 and x2 input features.
- Set a small network and record the data split, noise and learning settings.
- Run training and watch the output map change; record both training and test loss.
- Reset and change one complexity setting, such as the number of hidden neurons, keeping other choices fixed.
- Train again and compare the losses and decision boundaries, noting any random variation between runs.
You’ll have: A settings-and-results comparison with an explanation of the model’s visible boundary.
Try this example
Use this sample prompt or task input as a starting point. Adapt the subject and level to your learners.
Experiment task: classify the built-in circular dataset using x1 and x2. Compare a small hidden layer with a larger one while keeping the learning rate, noise and data split fixed. Which boundary fits the pattern, and what does test loss reveal?
Check your result before sharing
Discuss variation between resets and why low training loss may fail to generalize. The colors show this experiment’s model output, not a guarantee about new real-world data.
Use the built-in sample data for demonstration; this is not a tool for training arbitrary personal or media datasets.
Other activities to explore
- Demonstrate how a network changes its decision boundary during training.
- Compare training and test loss when changing model complexity.
Keep the official guide beside you
Menus and included features can vary by account or plan. Use the provider’s instructions if your screen differs.
This workflow is an editorial suggestion based on documentation, not a hands-on test or a claim of learning gains. Check school approval, current pricing, and student account eligibility before a classroom rollout. Use public or permitted samples and keep identifiable student information out of unapproved services.