public static void LoadData()
{
CsvDataFrame frame = new CsvDataFrame();
Downloader.DownloadSample(SampleDataset.HousingRegression);
var samplePath = Downloader.GetSamplePath(SampleDataset.HousingRegression);
frame.ReadCsv(samplePath.Train, true);
}
public static void LoadData()
{
CsvDataFrame <float> frame = new CsvDataFrame <float>();
Downloader.DownloadSample(SampleDataset.HousingRegression);
var samplePath = Downloader.GetSamplePath(SampleDataset.HousingRegression);
frame.ReadCsv(samplePath.Train, true);
DataFrame <float> trainX = frame[1, 13];
DataFrame <float> trainY = frame[14, 1];
trainData = new DataFrameList <float>(trainX, trainY);
}
static void Main(string[] args)
{
//Environment.SetEnvironmentVariable("MXNET_ENGINE_TYPE", "NaiveEngine");
mx.SetDevice(DeviceType.CPU);
//uint batchSize = 200;
//uint numFeatures = 13;
//var x = Symbol.Variable("x");
//var trainData = ReadCsv("./data/train.csv");
//var x_train = trainData.SliceAxis(1, 1, 14);
//var y_train = trainData.SliceAxis(1, 14, 15);
//NDArrayIter dataIter = new NDArrayIter(x_train, y_train);
//var fc1 = sym.FullyConnected(x, Symbol.Variable("fc1_w"), null, 64, no_bias: true, symbol_name: "fc1");
//var fc2 = sym.Relu(sym.FullyConnected(fc1, Symbol.Variable("fc2_w"), null, 32, no_bias: true, symbol_name: "fc2"), "relu2");
//var fc3 = sym.FullyConnected(fc2, Symbol.Variable("fc3_w"), null, 1, no_bias: true, symbol_name: "fc3");
//var output = sym.LinearRegressionOutput(fc3, Symbol.Variable("label"), symbol_name: "model");
//NDArrayDict parameters = new NDArrayDict();
//parameters["x"] = new NDArray(new Shape(batchSize, numFeatures));
//parameters["label"] = new NDArray(new Shape(batchSize));
//output.InferArgsMap(MXNet.Device, parameters, parameters);
//foreach (var item in parameters.ToList())
//{
// if (item.Key == "x" || item.Key == "label")
// continue;
// item.Value.SampleUniform();
//}
//var opt = new Adam();
//BaseMetric metric = new MAE();
//using (var exec = output.SimpleBind(MXNet.Device, parameters))
//{
// dataIter.SetBatch(batchSize);
// var argNames = output.ListArguments();
// DataBatch batch;
// for (int iter = 1; iter <= 1000; iter++)
// {
// dataIter.Reset();
// metric.Reset();
// while (dataIter.Next())
// {
// batch = dataIter.GetDataBatch();
// batch.Data.CopyTo(parameters["x"]);
// batch.Label.CopyTo(parameters["label"]);
// exec.Forward(true);
// exec.Backward();
// for (var i = 0; i < argNames.Count; ++i)
// {
// if (argNames[i] == "x" || argNames[i] == "label")
// continue;
// opt.Update(iter, i, exec.ArgmentArrays[i], exec.GradientArrays[i]);
// }
// metric.Update(parameters["label"], exec.Output);
// }
// Console.WriteLine("Iteration: {0}, Metric: {1}", iter, metric.Get());
// }
//}
//Global.Device = Context.Cpu();
////Read Data
CsvDataFrame trainReader = new CsvDataFrame("./data/train.csv", true);
trainReader.ReadCsv();
var trainX = trainReader[1, 14];
var trainY = trainReader[14, 15];
CsvDataFrame valReader = new CsvDataFrame("./data/test.csv", true);
valReader.ReadCsv();
var valX = valReader[1, 14];
NDArrayIter train = new NDArrayIter(trainX, trainY);
//Build Model
var model = new Module(13);
model.Add(new Dense(64, ActivationType.ReLU));
model.Add(new Dense(32, ActivationType.ReLU));
model.Add(new Dense(1));
model.Compile(OptimizerRegistry.Adam(), LossType.MeanSquaredError, new MSE());
model.Fit(train, 1000, 32);
Console.ReadLine();
}