Hands-On Machine Learning for Algorithmic Trading

Visualizing the encoding

We can use the t-distributed Stochastic Neighbor Embedding (t-SNE) manifold learning technique, see Chapter 12, Unsupervised Learning, to visualize and assess the quality of the encoding learned by the autoencoder's hidden layer.

If the encoding is successful in capturing the salient features of the data, the compressed representation of the data should still reveal a structure aligned with the 10 classes that differentiate the observations.

We use the output of the deep encoder we just trained to obtain the 32-dimensional representation of the test set:

tsne = TSNE(perplexity=25, n_iter=5000)train_embed = tsne.fit_transform(encoder_deep.predict(X_train_scaled))

The following figure shows that t-SNE manages to separate ...