How to get started with TensorFlow Playground

Google’s TensorFlow Playground is an interactive visualisation of neural networks. With it, you can simulate small neural networks in real-time in your browser, and see the results instantly.

Here, we we will train your first little neural network and help to familiarise you with the concept of artificial neural networks (ANNs). 

The concept of ANNs

The main purpose of this task is to explore a bit of what is happening inside the most famous machine learning algorithms and to understand how typical parameters influence learning.

The Task:

1 Go to playground.tensorflow.org

2 Settings
In the top part of the menu you’ll see the network hyperparameter (settings):

• Learning rate; how fast the network learns
• Activation function (a mathematic function – don’t worry, you can just try different ones if you like; ReLu learns fastest);
• Epoch; the number of times that the ANN will work through the entire training dataset
• Regularisation; ensures better training

3 Data

• Select the type of data you want to use from Data (on the left side) and the classification type problem
• NB. There are four types of classification and two types of regression (regression means that one does not just predict classes, but also real values, e.g. the colour intensity in these examples).

4 Increasing the number of neurons – Linear activation

• Try increasing the number of neurons in the hidden layer from 1 to 2, and also try changing from a linear activation to a nonlinear activation like ReLU.__

5 Increasing the number of neurons – Nonlinear activation

• Try increasing the number of neurons in the hidden layer from 2 to 3, using a nonlinear activation like ReLU.

6 Adding or removing hidden layers

• Continue experimenting by adding or removing hidden layers and neurons per layer. Feel free to change learning rates, regularisation and other learning settings.