Main Content


Class: Autoencoder

Convert Autoencoder object into network object



net = network(autoenc) returns a network object which is equivalent to the autoencoder, autoenc.

Input Arguments

expand all

Trained autoencoder, returned as an object of the Autoencoder class.

Output Arguments

expand all

Neural network, that is equivalent to the autoencoder autoenc, returned as an object of the network class.


expand all

Load the sample data.

X = bodyfat_dataset;

X = bodyfat_dataset;

X is a 13-by-252 matrix defining thirteen attributes of 252 different neighborhoods. For more information on the data, type help house_dataset in the command line.

Train an autoencoder on the attribute data.

autoenc = trainAutoencoder(X);

Create a network object from the autoencoder, autoenc .

net = network(autoenc);

Predict the attributes using the network, net .

Xpred = net(X);

Fit a linear regression model between the actual and estimated attributes data. Compute the estimated Pearson correlation coefficient, the slope and the intercept (bias) of the regression model, using all attribute data as one data set.

[C, S, B] = regression(X, Xpred, 'one')
C = 0.9997
S = 0.9984
B = 0.0960

The correlation coefficient is almost 1, which indicates that the attributes data and the estimations from the neural network are highly close to each other.

Plot the actual data and the fitted line.

plotregression(X, Xpred);

Figure Regression (plotregression) contains an axes. The axes with title : R=0.99969 contains 3 objects of type line. These objects represent Y = T, Fit, Data.

The data appears to be on the fitted line, which visually supports the conclusion that the predictions are very close to the actual data.

Introduced in R2015b