Early Stopping

What is early stopping?

When training neural networks, numerous decisions need to be made regarding the settings (hyperparameters) used, in order to obtain good performance. Once such hyperparameter is the number of training epochs: that is, how many full passes of the data set (epochs) should be used? If we use too few epochs, we might underfit (i.e., not learn everything we can from the training data); if we use too many epochs, we might overfit (i.e., fit the ‘noise’ in the training data, and not the signal).

Early stopping attempts to remove the need to manually set this value. It can also be considered a type of regularization method (like L1/L2 weight decay and dropout) in that it can stop the network from overfitting.

The idea behind early stopping is relatively simple:

  • Split data into training and test sets
  • At the end of each epoch (or, every N epochs):
    • evaluate the network performance on the test set
    • if the network outperforms the previous best model: save a copy of the network at the current epoch
  • Take as our final model the model that has the best test set performance

This is shown graphically below:

Early Stopping

The best model is the one saved at the time of the vertical dotted line - i.e., the model with the best accuracy on the test set.

Using DL4J’s early stopping functionality requires you to provide a number of configuration options:

  • A score calculator, such as the DataSetLossCalculator(JavaDoc, Source Code) for a Multi Layer Network, or DataSetLossCalculatorCG (JavaDoc, Source Code) for a Computation Graph. Is used to calculate at every epoch (for example: the loss function value on a test set, or the accuracy on the test set)
  • How frequently we want to calculate the score function (default: every epoch)
  • One or more termination conditions, which tell the training process when to stop. There are two classes of termination conditions:
    • Epoch termination conditions: evaluated every N epochs
    • Iteration termination conditions: evaluated once per minibatch
  • A model saver, that defines how models are saved

An example, with an epoch termination condition of maximum of 30 epochs, a maximum of 20 minutes training time, calculating the score every epoch, and saving the intermediate results to disk:

MultiLayerConfiguration myNetworkConfiguration = ...;
DataSetIterator myTrainData = ...;
DataSetIterator myTestData = ...;

EarlyStoppingConfiguration esConf = new EarlyStoppingConfiguration.Builder()
		.epochTerminationConditions(new MaxEpochsTerminationCondition(30))
		.iterationTerminationConditions(new MaxTimeIterationTerminationCondition(20, TimeUnit.MINUTES))
		.scoreCalculator(new DataSetLossCalculator(myTestData, true))
		.modelSaver(new LocalFileModelSaver(directory))

EarlyStoppingTrainer trainer = new EarlyStoppingTrainer(esConf,myNetworkConfiguration,myTrainData);

//Conduct early stopping training:
EarlyStoppingResult result = trainer.fit();

//Print out the results:
System.out.println("Termination reason: " + result.getTerminationReason());
System.out.println("Termination details: " + result.getTerminationDetails());
System.out.println("Total epochs: " + result.getTotalEpochs());
System.out.println("Best epoch number: " + result.getBestModelEpoch());
System.out.println("Score at best epoch: " + result.getBestModelScore());

//Get the best model:
MultiLayerNetwork bestModel = result.getBestModel();

You can also implement your own iteration and epoch termination conditions.

Early Stopping w/ Parallel Wrapper

The early stopping implementation described above will only work with a single device. However, EarlyStoppingParallelTrainer provides similar functionality as early stopping and allows you to optimize for either multiple CPUs or GPUs. EarlyStoppingParallelTrainer wraps your model in a ParallelWrapper class and performs localized distributed training.

Note that EarlyStoppingParallelTrainer doesn’t support all of the functionality as its single device counterpart. It is not UI-compatible and may not work with complex iteration listeners. This is due to how the model is distributed and copied in the background.




Score function for a MultiLayerNetwork or ComputationGraph with a single



as accuracy, F1 score, etc. Used for both MultiLayerNetwork and ComputationGraph



Calculate the score (loss function value) on a given data set (usually a test set)



Given a DataSetIterator: calculate the total loss for the model on that data set. Typically used to calculate the loss on a test set.



Calculate ROC AUC (area under ROC curve) or AUCPR (area under precision recall curve) for a MultiLayerNetwork or ComputationGraph



Calculate the regression score of the network (MultiLayerNetwork or ComputationGraph) on a test set, using the



Score function for variational autoencoder reconstruction error for a MultiLayerNetwork or ComputationGraph.
VariationalAutoencoder layer must be first layer in the network

public boolean minimizeScore() 

Constructor for reconstruction ERROR

  • param metric
  • param iterator



Score calculator for variational autoencoder reconstruction probability or reconstruction log probability for a MultiLayerNetwork or ComputationGraph. VariationalAutoencoder layer must be first layer in the network

public boolean minimizeScore() 

Constructor for average reconstruction probability

  • param iterator Iterator
  • param reconstructionProbNumSamples Number of samples. See {- link VariationalAutoencoder#reconstructionProbability(INDArray, int)} for details
  • param logProb If true: calculate (negative) log probability. False: probability



Created by Sadat Anwar on 3/26/16.

Stop the training once we achieved an expected score. Normally this will stop if the current score is lower than the initialized score. If you want to stop the training once the score increases the defined score set the lesserBetter flag to false (feel free to give the flag a better name)

public void initialize() 
  • deprecated “lessBetter” argument no longer used



Terminate training if best model score does not improve for N epochs

API Reference

API Reference

Detailed API docs for all libraries including DL4J, ND4J, DataVec, and Arbiter.



Explore sample projects and demos for DL4J, ND4J, and DataVec in multiple languages including Java and Kotlin.



Step-by-step tutorials for learning concepts in deep learning while using the DL4J API.



In-depth documentation on different scenarios including import, distributed training, early stopping, and GPU setup.

Subscribe to the DL4J Newsletter