Package ai.djl.training.hyperparameter

Class EasyHpo

java.lang.Object

ai.djl.training.hyperparameter.EasyHpo

public abstract class EasyHpo
extends java.lang.Object

Helper for easy training with hyperparameters.

Constructor Summary

Constructors

Constructor	Description
EasyHpo()

Method Summary

Modifier and Type	Method	Description
protected abstract Model	buildModel(HpSet hpVals)	Builds the Model and Block to train.
ai.djl.util.Pair<Model,TrainingResult>	fit()	Fits the model given the implemented abstract methods.
protected abstract RandomAccessDataset	getDataset(Dataset.Usage usage)	Returns the dataset to train with.
protected abstract Shape	inputShape(HpSet hpVals)	Returns the input shape for the model.
protected abstract int	numEpochs(HpSet hpVals)	Returns the number of epochs to train for the current hyperparameter set.
protected abstract int	numHyperParameterTests()	Returns the number of hyperparameter sets to train with.
protected void	saveModel(Model model, TrainingResult result)	Saves the best hyperparameter set.
protected abstract HpSet	setupHyperParams()	Returns the initial hyperparameters.
protected abstract TrainingConfig	setupTrainingConfig(HpSet hpVals)	Returns the TrainingConfig to use to train each hyperparameter set.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

EasyHpo

public EasyHpo()

Method Detail

fit

public ai.djl.util.Pair<Model,TrainingResult> fit()
                                                 throws java.io.IOException,
                                                        TranslateException

Fits the model given the implemented abstract methods.

Returns:

the best model and training results

Throws:

java.io.IOException - for various exceptions depending on the dataset

TranslateException - if there is an error while processing input

setupHyperParams

protected abstract HpSet setupHyperParams()

Returns the initial hyperparameters.

Returns:

the initial hyperparameters

getDataset

protected abstract RandomAccessDataset getDataset(Dataset.Usage usage)
                                           throws java.io.IOException

Returns the dataset to train with.

Parameters:

usage - the usage of the dataset

Returns:

the dataset to train with

Throws:

java.io.IOException - if the dataset could not be loaded

setupTrainingConfig

protected abstract TrainingConfig setupTrainingConfig(HpSet hpVals)

Returns the TrainingConfig to use to train each hyperparameter set.

Parameters:

hpVals - the hyperparameters to train with

Returns:

the TrainingConfig to use to train each hyperparameter set

buildModel

protected abstract Model buildModel(HpSet hpVals)

Builds the Model and Block to train.

Parameters:

hpVals - the hyperparameter values to use for the model

Returns:

the model to train

inputShape

protected abstract Shape inputShape(HpSet hpVals)

Returns the input shape for the model.

Parameters:

hpVals - the hyperparameter values for the model

Returns:

returns the model input shape

numEpochs

protected abstract int numEpochs(HpSet hpVals)

Returns the number of epochs to train for the current hyperparameter set.

Parameters:

hpVals - the current hyperparameter set

Returns:

the number of epochs

numHyperParameterTests

protected abstract int numHyperParameterTests()

Returns the number of hyperparameter sets to train with.

Returns:

the number of hyperparameter sets to train with

saveModel

protected void saveModel(Model model,
                         TrainingResult result)
                  throws java.io.IOException

Saves the best hyperparameter set.

Parameters:

model - the model to save

result - the training result for training with this model's hyperparameters

Throws:

java.io.IOException - if the model could not be saved