Method Fit

model/model.go:419–697

(
	dataset data.Dataset,
	config FitConfig,
)

}

func (m *TfkgModel) Fit(
	dataset data.Dataset,
	config FitConfig,
) {
	trainSig := m.model.Signatures["learn"]
	var trainOutputs []tf.Output
	output := 0
	for _, info := range trainSig.Outputs {
		parts := strings.Split(info.Name, ":")
		if len(parts) != 2 {
			e := fmt.Errorf("error getting output for train signature in fit")
			m.errorHandler.Error(e)
			return
		}
		name := parts[0]
		trainOutputs = append(trainOutputs, m.model.Graph.Operation(name).Output(output))
		output++
	}

	if config.Epochs == 0 {
		config.Epochs = 1
	}

	for i := range config.Metrics {
		config.Metrics[i].Init()
	}
	for i := range config.Callbacks {
		e := config.Callbacks[i].Init()
		if e != nil {
			m.errorHandler.Error(e)
			return
		}
	}

	for epoch := 1; epoch <= config.Epochs; epoch++ {

		generatorChan := dataset.
			SetMode(data.GeneratorModeTrain).
			GeneratorChan(config.BatchSize, config.PreFetch)

		labelOp := m.model.Graph.Operation(fmt.Sprintf("learn_%s", "y")).Output(0)
		classWeightOp := m.model.Graph.Operation(fmt.Sprintf("learn_%s", "class_weights")).Output(0)

		var inputOps []tf.Output
		for offset := range dataset.GetColumnNames() {
			inputOps = append(inputOps, m.model.Graph.Operation(fmt.Sprintf("learn_inputs_%d", offset)).Output(0))
		}

		halt := false

		var trainLogs []callback.Log

		batch := 1
		totalBatches := dataset.Len() / config.BatchSize
		trainTotalLoss := float64(0)
		swg := sizedwaitgroup.New(runtime.NumCPU())
		modelLock := &sync.Mutex{}
		for generatorBatch := range generatorChan {