Trainer (Deep Java Library 0.5.0 API specification)

java.lang.Object
- ai.djl.training.Trainer

All Implemented Interfaces:

java.lang.AutoCloseable
```
public class Trainer
extends java.lang.Object
implements java.lang.AutoCloseable
```
The Trainer interface provides a session for model training.
Trainer provides an easy, and manageable interface for training. Trainer is not thread-safe.
See the tutorials on:

Constructor Summary

Constructors
Constructor and Description

Trainer(Model model, TrainingConfig trainingConfig)
Creates an instance of Trainer with the given Model and TrainingConfig.

Constructors
Constructor and Description
`Trainer(Model model, TrainingConfig trainingConfig)` Creates an instance of `Trainer` with the given `Model` and `TrainingConfig`.

Method Summary

All Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`void`	`close()`
`void`	`endEpoch()` Runs the end epoch actions.
`protected void`	`finalize()`
`NDList`	`forward(NDList input)` Applies the forward function of the model once on the given input `NDList`.
`java.util.List<Device>`	`getDevices()` Returns the devices used for training.
`java.util.List<Evaluator>`	`getEvaluators()` Gets all `Evaluator`s.
`Loss`	`getLoss()` Gets the training `Loss` function of the trainer.
`NDManager`	`getManager()` Gets the `NDManager` from the model.
`Metrics`	`getMetrics()` Returns the Metrics param used for benchmarking.
`Model`	`getModel()` Returns the model used to create this trainer.
`TrainingResult`	`getTrainingResult()` Returns the `TrainingResult`.
`void`	`initialize(Shape... shapes)` Initializes the `Model` that the `Trainer` is going to train.
`java.lang.Iterable<Batch>`	`iterateDataset(Dataset dataset)` Fetches an iterator that can iterate through the given `Dataset`.
`GradientCollector`	`newGradientCollector()` Returns a new instance of `GradientCollector`.
`NDList`	`predict(NDList input)` Applies the predict function of the model once on the given input `NDList`.
`void`	`setMetrics(Metrics metrics)` Attaches a Metrics param to use for benchmarking.
`void`	`step()` Updates all of the parameters of the model once.
`void`	`trainBatch(Batch batch)` Trains the model with one iteration of the given `Batch` of data.
`void`	`validateBatch(Batch batch)` Validates the given batch of data.

Methods inherited from class java.lang.Object
clone, equals, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

- Constructor Detail
  - Trainer
```
public Trainer(Model model,
               TrainingConfig trainingConfig)
```
    Creates an instance of Trainer with the given Model and TrainingConfig.
    
    Parameters:
    
    model - the model the trainer will train on
    
    trainingConfig - the configuration used by the trainer
- Method Detail
  - initialize
```
public void initialize(Shape... shapes)
```
    Initializes the Model that the Trainer is going to train.
    
    Parameters:
    
    shapes - an array of Shape of the inputs
  - iterateDataset
```
public java.lang.Iterable<Batch> iterateDataset(Dataset dataset)
```
    Fetches an iterator that can iterate through the given Dataset.
    
    Parameters:
    
    dataset - the dataset to iterate through
    
    Returns:
    
    an Iterable of Batch that contains batches of data from the dataset
  - newGradientCollector
```
public GradientCollector newGradientCollector()
```
    Returns a new instance of GradientCollector.
    
    Returns:
    
    a new instance of GradientCollector
  - trainBatch
```
public void trainBatch(Batch batch)
```
    Trains the model with one iteration of the given Batch of data.
    
    Parameters:
    
    batch - a Batch that contains data, and its respective labels
    
    Throws:
    
    java.lang.IllegalArgumentException - if the batch engine does not match the trainer engine
  - forward
```
public NDList forward(NDList input)
```
    Applies the forward function of the model once on the given input NDList.
    
    Parameters:
    
    input - the input NDList
    
    Returns:
    
    the output of the forward function
  - predict
```
public NDList predict(NDList input)
```
    Applies the predict function of the model once on the given input NDList.
    
    Parameters:
    
    input - the input NDList
    
    Returns:
    
    the output of the predict function
  - validateBatch
```
public void validateBatch(Batch batch)
```
    Validates the given batch of data.
    During validation, the evaluators and losses are computed, but gradients aren't computed, and parameters aren't updated.
    
    Parameters:
    
    batch - a Batch of data
    
    Throws:
    
    java.lang.IllegalArgumentException - if the batch engine does not match the trainer engine
  - step
```
public void step()
```
    Updates all of the parameters of the model once.
  - getMetrics
```
public Metrics getMetrics()
```
    Returns the Metrics param used for benchmarking.
    
    Returns:
    
    the the Metrics param used for benchmarking
  - setMetrics
```
public void setMetrics(Metrics metrics)
```
    Attaches a Metrics param to use for benchmarking.
    
    Parameters:
    
    metrics - the Metrics class
  - getDevices
```
public java.util.List<Device> getDevices()
```
    Returns the devices used for training.
    
    Returns:
    
    the devices used for training
  - endEpoch
```
public void endEpoch()
```
    Runs the end epoch actions.
  - getLoss
```
public Loss getLoss()
```
    Gets the training Loss function of the trainer.
    
    Returns:
    
    the Loss function
  - getModel
```
public Model getModel()
```
    Returns the model used to create this trainer.
    
    Returns:
    
    the model associated with this trainer
  - getEvaluators
```
public java.util.List<Evaluator> getEvaluators()
```
    Gets all Evaluators.
    
    Returns:
    
    the evaluators used during training
  - getTrainingResult
```
public TrainingResult getTrainingResult()
```
    Returns the TrainingResult.
    
    Returns:
    
    the TrainingResult
  - getManager
```
public NDManager getManager()
```
    Gets the NDManager from the model.
    
    Returns:
    
    the NDManager
  - finalize
```
protected void finalize()
                 throws java.lang.Throwable
```
    Overrides:
    
    finalize in class java.lang.Object
    
    Throws:
    
    java.lang.Throwable
  - close
```
public void close()
```
    Specified by:
    
    close in interface java.lang.AutoCloseable

Class Trainer

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Constructor Detail

Trainer

Method Detail

initialize

iterateDataset

newGradientCollector

trainBatch

forward

predict

validateBatch

step

getMetrics

setMetrics

getDevices

endEpoch

getLoss

getModel

getEvaluators

getTrainingResult

getManager

finalize

close