static void Main(string[] args)
{
// unpack archive
Console.WriteLine("Loading data...");
if (!System.IO.File.Exists("x_train.bin"))
{
DataUtil.Unzip(@"..\..\..\..\..\imdb_data.zip", ".");
}
// load training and test data
var training_data = DataUtil.LoadBinary <float>("x_train.bin", 25000, 10000);
var training_labels = DataUtil.LoadBinary <float>("y_train.bin", 25000);
var test_data = DataUtil.LoadBinary <float>("x_test.bin", 25000, 10000);
var test_labels = DataUtil.LoadBinary <float>("y_test.bin", 25000);
// create feature and label variables
var features = NetUtil.Var(new[] { 10000 }, DataType.Float);
var labels = NetUtil.Var(new[] { 1 }, DataType.Float);
// create neural network
var network = features
.Dense(16, CNTKLib.ReLU)
.Dense(16, CNTKLib.ReLU)
.Dense(1, CNTKLib.Sigmoid)
.ToNetwork();
// create loss and test functions
var lossFunc = CNTKLib.BinaryCrossEntropy(network.Output, labels);
var accuracyFunc = NetUtil.BinaryAccuracy(network.Output, labels);
// use the Adam learning algorithm
var learner = network.GetAdamLearner(
learningRateSchedule: (0.001, 1),
momentumSchedule: (0.9, 1),
unitGain: true);
// get a trainer for training, and an evaluator for testing the network
var trainer = network.GetTrainer(learner, lossFunc, accuracyFunc);
var evaluator = network.GetEvaluator(accuracyFunc);
// declare some variables
var trainingError = new List <double>();
var validationError = new List <double>();
var maxEpochs = 7;
var batchSize = 32;
var batchCount = 0;
var error = 0.0;
// train for a number of epochs
Console.WriteLine("Training network...");
for (int epoch = 0; epoch < maxEpochs; epoch++)
{
error = 0.0;
batchCount = 0;
// train the network using batches
training_data.Index().Shuffle().Batch(
batchSize,
(indices, begin, end) =>
{
// get the current batch for training
var featureBatch = features.GetBatch(training_data, indices, begin, end);
var labelBatch = labels.GetBatch(training_labels, indices, begin, end);
// train the network on the batch
var result = trainer.TrainBatch(
new[]
{
(features, featureBatch),
(labels, labelBatch)
},
