WOLFRAM

Wolfram Language & System Documentation Center
Regression with Neural Networks

Regression with Neural Networks

Regression from Numeric Data
Train a predictor that predicts the median value of properties in a neighborhood of Boston, given some features of the neighborhood.
First, obtain the training and validation data:
Wolfram Language code: train = ExampleData[{"MachineLearning", "BostonHomes"}, "TrainingData"]; test = ExampleData[{"MachineLearning", "BostonHomes"}, "TestData"];
Wolfram Language code: First[train]
Define a net that returns a scalar:
Wolfram Language code: net = NetChain[{LinearLayer[15], BatchNormalizationLayer[], ElementwiseLayer[Ramp], LinearLayer[10], BatchNormalizationLayer[], ElementwiseLayer[Ramp], LinearLayer[1], PartLayer[1]}, "Input" -> 13 ]
Train the net and determine the lowest validation loss:
Wolfram Language code: results = NetTrain[net, train, All, ValidationSet -> test] {trainedNet, valLosses} = results[{"TrainedNet", "ValidationLossList"}]; Min[valLosses]
Make a prediction:
Wolfram Language code: trainedNet[{0.02731, 0, 7.07, 0, 0.469, 6.421, 78.9, 4.9671, 2, 242, 17.8, 396.9, 9.14}]
Train a predictor using Predict instead, and obtain the mean squared loss on the test set:
Wolfram Language code: p = Predict[train, PerformanceGoal -> "Quality"] pm = PredictorMeasurements[p, test, "MeanSquare"]