Fundamentals of Machine Learning for Software Engineers/

...

/

Keras in Action

Keras in Action

Build neural network model using Keras.

We'll cover the following...

Implementation of neural network using Keras

Let’s move on to the code as we have discussed each component in detail already. Here is the whole thing.

To save time, a decision boundary is created after 1000 epochs. However to improve the decision boundary, epochs must be increased as shown after the code widget for 30,000 epochs.

Please wait for 1 minute after the training phase completes and then click the app link to preview the decision boundary:

<!DOCTYPE html>
<html>
    <body>
        <center>
        <img src="graph.png" alt="output" width="600" height="400">
        </center>
    </body>
</html>
Neural network using Keras

Here’s the output after training the network for 30,000 epochs, stripped down to the essential information:

Using TensorFlow back-end.
Train on 285 samples, validate on 285 samples
Epoch 1 - loss: 0.7222 - acc: 0.5088 - val_loss: 0.6928 - val_acc: 0.4807
Epoch 2 - loss: 0.6966 - acc: 0.5018 - val_loss: 0.6904 - val_acc: 0.5193
…
Epoch 30000 - loss: 0.1623 - acc: 0.9193 - val_loss: 0.1975 - val_acc: 0.9018

The accuracy on the training set gets up to ...