engine

Documentation

Overview

Package engine defines the Engine interface.

Index

• type Engine

Types

type Engine

type Engine interface {

	// Number of dimensions in the predictions. Predictions (allocated by
	// "AllocatePredictions" and populated by "Predict") contains "OutputDim"
	// elements for each example (example major; output dim minor).
	OutputDim() int

	// Populates "predictions" with the predictions of the engine.
	// Is "numExamples" is less than the number of examples allocated in
	// "examples", "Predict" computes the predictions for the first "numExamples"
	// examples is computed.
	Predict(examples *example.Batch, numExamples int, predictions []float32)

	// Allocates a set of examples. A same batch can be re-used multiple time.
	AllocateExamples(maxNumExamples int) *example.Batch

	// Allocates a set of predictions.
	AllocatePredictions(maxNumExamples int) []float32

	// Input features of the model. "Features()" is used to generated the feature
	// ids (e.g. "Features().GetNumericalFeatureID("age")") of the model during
	// the initialization phase.
	Features() *example.Features
}

Engine generates predictions for a model, given an example.

Usage example:

import (".../ports/go/model/io")

// Load a model
model, err := io.Load("/path/to/model")

// Compile the model for fast infernce.
// At this point, the "model objet can be discarded.
engine, err := NewEngine(model)

// b/235436797: When running a model trained with the TensorFlow Decision
// Forests API, use the `NewEngineWithCompatibility` method instead:
engine, err := serving.NewEngineWithCompatibility(model, example.CompatibilityTensorFlowDecisionForests)

// Obtain the ID of the model input feature (should be only done once).
feature_age, age_used_by_model := engine.Features().NumericalFeatures["age"]
feature_country, country_used_by_model := engine.Features().CategoricalFeatures["country"]

// Allocate a batch of examples and predictions
examples := engine.AllocateExamples(10)
predictions := engine.AllocatePredictions(10)

// Clear the content of examples.
// This is only useful if "examples" is re-used multiple times.
examples.Clear()

// Set all the example values as missing. This operation is only necessary if
// not all the feature values will be set.
examples.FillMissing()

// Set the feature "age" to be 32 for the first examples.
examples.SetNumerical(, feature_age, 32.0)

// Set the feature "age" to be missing for the second examples.
examples.SetMissingNumerical(, feature_age)

// Set the feature "country" to be "UK" for the first examples.
examples.SetCategorical(, feature_country, "UK")

// Generates the predictions for the first two examples.
engine.Predict(examples, 2, predictions)

// Print the predictions.
fmt.Println("Each prediction is of dimension %v", engine.OutputDim())
fmt.Println("The predictions are %v and %v",predictions[0], predictions[1])