Function main

examples/bench/main.go:21–146

()

)

func main() {
	// This is where the trained model will be saved
	saveDir := filepath.Join("../../logs", fmt.Sprintf("bench-%d", time.Now().Unix()))
	e := os.MkdirAll(saveDir, os.ModePerm)
	if e != nil {
		panic(e)
	}

	// Create a logger pointed at the save dir
	logger, e := cblog.NewLogger(cblog.LoggerConfig{
		LogLevel:           cblog.DebugLevel,
		Format:             "%{time:2006-01-02 15:04:05.000} : %{file}:%{line} : %{message}",
		LogToFile:          true,
		FilePath:           filepath.Join(saveDir, "training.log"),
		FilePerm:           os.ModePerm,
		LogToStdOut:        true,
		SetAsDefaultLogger: true,
	})
	if e != nil {
		panic(e)
	}

	// Error handler with stack traces
	errorHandler := cberrors.NewErrorContainer(iowriterprovider.New(logger))

	// Create a new Values dataset, we can pass in values without having to read from a CSV file
	dataset, e := data.NewValuesDataset(
		logger,
		errorHandler,
		data.ValuesDatasetConfig{
			TrainPercent: 1,
		},
		preprocessor.NewProcessor(
			errorHandler,
			"y",
			preprocessor.ProcessorConfig{
				Converter: preprocessor.ConvertInterfaceToInt32SliceTensor,
			},
		),
		preprocessor.NewProcessor(
			errorHandler,
			"floats",
			preprocessor.ProcessorConfig{
				Converter: preprocessor.ConvertInterfaceFloat32SliceToTensor,
			},
		),
	)
	if e != nil {
		errorHandler.Error(e)
		return
	}

	var y []interface{}
	var x []interface{}
	// generate random input data with imbalanced classes, if class_weighting works the accuracy should be 0
	for i := 0; i < 29000; i++ {
		var xRow []float32
		for j := 0; j < 1000; j++ {