DeepLearningParametersV3 (h2o-bindings version 3.18.0.6 API)

java.lang.Object
- water.bindings.pojos.SchemaV3
- - water.bindings.pojos.ModelParametersSchemaV3
  - - water.bindings.pojos.DeepLearningParametersV3

public class DeepLearningParametersV3
extends ModelParametersSchemaV3

Field Summary

Fields
Modifier and Type	Field and Description
`DeepLearningActivation`	`activation` Activation function.
`boolean`	`adaptiveRate` Adaptive learning rate.
`boolean`	`autoencoder` Auto-Encoder.
`double`	`averageActivation` Average activation for sparse auto-encoder.
`boolean`	`balanceClasses` Balance training data class counts via over/under-sampling (for imbalanced data).
`double`	`classificationStop` Stopping criterion for classification error fraction on training data (-1 to disable).
`float[]`	`classSamplingFactors` Desired over/under-sampling ratios per class (in lexicographic order).
`boolean`	`colMajor` #DEPRECATED Use a column major weight matrix for input layer.
`boolean`	`diagnostics` Enable diagnostics for hidden layers.
`boolean`	`elasticAveraging` Elastic averaging between compute nodes can improve distributed model convergence.
`double`	`elasticAveragingMovingRate` Elastic averaging moving rate (only if elastic averaging is enabled).
`double`	`elasticAveragingRegularization` Elastic averaging regularization strength (only if elastic averaging is enabled).
`double`	`epochs` How many times the dataset should be iterated (streamed), can be fractional.
`double`	`epsilon` Adaptive learning rate smoothing factor (to avoid divisions by zero and allow progress).
`boolean`	`exportWeightsAndBiases` Whether to export Neural Network weights and biases to H2O Frames.
`boolean`	`fastMode` Enable fast mode (minor approximation in back-propagation).
`boolean`	`forceLoadBalance` Force extra load balancing to increase training speed for small datasets (to keep all cores busy).
`int[]`	`hidden` Hidden layer sizes (e.g.
`double[]`	`hiddenDropoutRatios` Hidden layer dropout ratios (can improve generalization), specify one value per hidden layer, defaults to 0.5.
`FrameKeyV3[]`	`initialBiases` A list of H2OFrame ids to initialize the bias vectors of this model with.
`DeepLearningInitialWeightDistribution`	`initialWeightDistribution` Initial weight distribution.
`FrameKeyV3[]`	`initialWeights` A list of H2OFrame ids to initialize the weight matrices of this model with.
`double`	`initialWeightScale` Uniform: -value...value, Normal: stddev.
`double`	`inputDropoutRatio` Input layer dropout ratio (can improve generalization, try 0.1 or 0.2).
`double`	`l1` L1 regularization (can add stability and improve generalization, causes many weights to become 0).
`double`	`l2` L2 regularization (can add stability and improve generalization, causes many weights to be small.
`DeepLearningLoss`	`loss` Loss function.
`float`	`maxAfterBalanceSize` Maximum relative size of the training data after balancing class counts (can be less than 1.0).
`int`	`maxCategoricalFeatures` Max.
`int`	`maxConfusionMatrixSize` [Deprecated] Maximum size (# classes) for confusion matrices to be printed in the Logs.
`int`	`maxHitRatioK` Max.
`float`	`maxW2` Constraint for squared sum of incoming weights per unit (e.g.
`int`	`miniBatchSize` Mini-batch size (smaller leads to better fit, larger can speed up and generalize better).
`DeepLearningMissingValuesHandling`	`missingValuesHandling` Handling of missing values.
`double`	`momentumRamp` Number of training samples for which momentum increases.
`double`	`momentumStable` Final momentum after the ramp is over (try 0.99).
`double`	`momentumStart` Initial momentum at the beginning of training (try 0.5).
`boolean`	`nesterovAcceleratedGradient` Use Nesterov accelerated gradient (recommended).
`boolean`	`overwriteWithBestModel` If enabled, override the final model with the best model found during training.
`ModelKeyV3`	`pretrainedAutoencoder` Pretrained autoencoder model to initialize this model with.
`boolean`	`quietMode` Enable quiet mode for less output to standard output.
`double`	`rate` Learning rate (higher => less stable, lower => slower convergence).
`double`	`rateAnnealing` Learning rate annealing: rate / (1 + rate_annealing * samples).
`double`	`rateDecay` Learning rate decay factor between layers (N-th layer: rate * rate_decay ^ (n - 1).
`double`	`regressionStop` Stopping criterion for regression error (MSE) on training data (-1 to disable).
`boolean`	`replicateTrainingData` Replicate the entire training dataset onto every node for faster training on small datasets.
`boolean`	`reproducible` Force reproducibility on small data (will be slow - only uses 1 thread).
`double`	`rho` Adaptive learning rate time decay factor (similarity to prior updates).
`double`	`scoreDutyCycle` Maximum duty cycle fraction for scoring (lower: more training, higher: more scoring).
`double`	`scoreInterval` Shortest time interval (in seconds) between model scoring.
`long`	`scoreTrainingSamples` Number of training set samples for scoring (0 for all).
`long`	`scoreValidationSamples` Number of validation set samples for scoring (0 for all).
`DeepLearningClassSamplingMethod`	`scoreValidationSampling` Method used to sample validation dataset for scoring.
`long`	`seed` Seed for random numbers (affects sampling) - Note: only reproducible when running single threaded.
`boolean`	`shuffleTrainingData` Enable shuffling of training data (recommended if training data is replicated and train_samples_per_iteration is close to #nodes x #rows, of if using balance_classes).
`boolean`	`singleNodeMode` Run on a single node for fine-tuning of model parameters.
`boolean`	`sparse` Sparse data handling (more efficient for data with lots of 0 values).
`double`	`sparsityBeta` Sparsity regularization.
`boolean`	`standardize` If enabled, automatically standardize the data.
`double`	`targetRatioCommToComp` Target ratio of communication overhead to computation.
`long`	`trainSamplesPerIteration` Number of training samples (globally) per MapReduce iteration.
`boolean`	`useAllFactorLevels` Use all factor levels of categorical variables.
`boolean`	`variableImportances` Compute variable importances for input features (Gedeon method) - can be slow for large networks.

Constructor Summary

Constructors
Constructor and Description

DeepLearningParametersV3()
Public constructor

Constructors
Constructor and Description
`DeepLearningParametersV3()` Public constructor

Method Summary

All Methods Instance Methods Concrete Methods
Modifier and Type Method and Description

java.lang.String toString()
Return the contents of this object as a JSON String.
- Methods inherited from class java.lang.Object
  clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

All Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`java.lang.String`	`toString()` Return the contents of this object as a JSON String.

- Field Detail
  - balanceClasses
```
public boolean balanceClasses
```
    Balance training data class counts via over/under-sampling (for imbalanced data).
  - classSamplingFactors
```
public float[] classSamplingFactors
```
    Desired over/under-sampling ratios per class (in lexicographic order). If not specified, sampling factors will be automatically computed to obtain class balance during training. Requires balance_classes.
  - maxAfterBalanceSize
```
public float maxAfterBalanceSize
```
    Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes.
  - maxConfusionMatrixSize
```
public int maxConfusionMatrixSize
```
    [Deprecated] Maximum size (# classes) for confusion matrices to be printed in the Logs.
  - maxHitRatioK
```
public int maxHitRatioK
```
    Max. number (top K) of predictions to use for hit ratio computation (for multi-class only, 0 to disable).
  - activation
```
public DeepLearningActivation activation
```
    Activation function.
  - hidden
```
public int[] hidden
```
    Hidden layer sizes (e.g. [100, 100]).
  - epochs
```
public double epochs
```
    How many times the dataset should be iterated (streamed), can be fractional.
  - trainSamplesPerIteration
```
public long trainSamplesPerIteration
```
    Number of training samples (globally) per MapReduce iteration. Special values are 0: one epoch, -1: all available data (e.g., replicated training data), -2: automatic.
  - targetRatioCommToComp
```
public double targetRatioCommToComp
```
    Target ratio of communication overhead to computation. Only for multi-node operation and train_samples_per_iteration = -2 (auto-tuning).
  - seed
```
public long seed
```
    Seed for random numbers (affects sampling) - Note: only reproducible when running single threaded.
  - adaptiveRate
```
public boolean adaptiveRate
```
    Adaptive learning rate.
  - rho
```
public double rho
```
    Adaptive learning rate time decay factor (similarity to prior updates).
  - epsilon
```
public double epsilon
```
    Adaptive learning rate smoothing factor (to avoid divisions by zero and allow progress).
  - rate
```
public double rate
```
    Learning rate (higher => less stable, lower => slower convergence).
  - rateAnnealing
```
public double rateAnnealing
```
    Learning rate annealing: rate / (1 + rate_annealing * samples).
  - rateDecay
```
public double rateDecay
```
    Learning rate decay factor between layers (N-th layer: rate * rate_decay ^ (n - 1).
  - momentumStart
```
public double momentumStart
```
    Initial momentum at the beginning of training (try 0.5).
  - momentumRamp
```
public double momentumRamp
```
    Number of training samples for which momentum increases.
  - momentumStable
```
public double momentumStable
```
    Final momentum after the ramp is over (try 0.99).
  - nesterovAcceleratedGradient
```
public boolean nesterovAcceleratedGradient
```
    Use Nesterov accelerated gradient (recommended).
  - inputDropoutRatio
```
public double inputDropoutRatio
```
    Input layer dropout ratio (can improve generalization, try 0.1 or 0.2).
  - hiddenDropoutRatios
```
public double[] hiddenDropoutRatios
```
    Hidden layer dropout ratios (can improve generalization), specify one value per hidden layer, defaults to 0.5.
  - l1
```
public double l1
```
    L1 regularization (can add stability and improve generalization, causes many weights to become 0).
  - l2
```
public double l2
```
    L2 regularization (can add stability and improve generalization, causes many weights to be small.
  - maxW2
```
public float maxW2
```
    Constraint for squared sum of incoming weights per unit (e.g. for Rectifier).
  - initialWeightDistribution
```
public DeepLearningInitialWeightDistribution initialWeightDistribution
```
    Initial weight distribution.
  - initialWeightScale
```
public double initialWeightScale
```
    Uniform: -value...value, Normal: stddev.
  - initialWeights
```
public FrameKeyV3[] initialWeights
```
    A list of H2OFrame ids to initialize the weight matrices of this model with.
  - initialBiases
```
public FrameKeyV3[] initialBiases
```
    A list of H2OFrame ids to initialize the bias vectors of this model with.
  - loss
```
public DeepLearningLoss loss
```
    Loss function.
  - scoreInterval
```
public double scoreInterval
```
    Shortest time interval (in seconds) between model scoring.
  - scoreTrainingSamples
```
public long scoreTrainingSamples
```
    Number of training set samples for scoring (0 for all).
  - scoreValidationSamples
```
public long scoreValidationSamples
```
    Number of validation set samples for scoring (0 for all).
  - scoreDutyCycle
```
public double scoreDutyCycle
```
    Maximum duty cycle fraction for scoring (lower: more training, higher: more scoring).
  - classificationStop
```
public double classificationStop
```
    Stopping criterion for classification error fraction on training data (-1 to disable).
  - regressionStop
```
public double regressionStop
```
    Stopping criterion for regression error (MSE) on training data (-1 to disable).
  - quietMode
```
public boolean quietMode
```
    Enable quiet mode for less output to standard output.
  - scoreValidationSampling
```
public DeepLearningClassSamplingMethod scoreValidationSampling
```
    Method used to sample validation dataset for scoring.
  - overwriteWithBestModel
```
public boolean overwriteWithBestModel
```
    If enabled, override the final model with the best model found during training.
  - autoencoder
```
public boolean autoencoder
```
    Auto-Encoder.
  - useAllFactorLevels
```
public boolean useAllFactorLevels
```
    Use all factor levels of categorical variables. Otherwise, the first factor level is omitted (without loss of accuracy). Useful for variable importances and auto-enabled for autoencoder.
  - standardize
```
public boolean standardize
```
    If enabled, automatically standardize the data. If disabled, the user must provide properly scaled input data.
  - diagnostics
```
public boolean diagnostics
```
    Enable diagnostics for hidden layers.
  - variableImportances
```
public boolean variableImportances
```
    Compute variable importances for input features (Gedeon method) - can be slow for large networks.
  - fastMode
```
public boolean fastMode
```
    Enable fast mode (minor approximation in back-propagation).
  - forceLoadBalance
```
public boolean forceLoadBalance
```
    Force extra load balancing to increase training speed for small datasets (to keep all cores busy).
  - replicateTrainingData
```
public boolean replicateTrainingData
```
    Replicate the entire training dataset onto every node for faster training on small datasets.
  - singleNodeMode
```
public boolean singleNodeMode
```
    Run on a single node for fine-tuning of model parameters.
  - shuffleTrainingData
```
public boolean shuffleTrainingData
```
    Enable shuffling of training data (recommended if training data is replicated and train_samples_per_iteration is close to #nodes x #rows, of if using balance_classes).
  - missingValuesHandling
```
public DeepLearningMissingValuesHandling missingValuesHandling
```
    Handling of missing values. Either MeanImputation or Skip.
  - sparse
```
public boolean sparse
```
    Sparse data handling (more efficient for data with lots of 0 values).
  - colMajor
```
public boolean colMajor
```
    #DEPRECATED Use a column major weight matrix for input layer. Can speed up forward propagation, but might slow down backpropagation.
  - averageActivation
```
public double averageActivation
```
    Average activation for sparse auto-encoder. #Experimental
  - sparsityBeta
```
public double sparsityBeta
```
    Sparsity regularization. #Experimental
  - maxCategoricalFeatures
```
public int maxCategoricalFeatures
```
    Max. number of categorical features, enforced via hashing. #Experimental
  - reproducible
```
public boolean reproducible
```
    Force reproducibility on small data (will be slow - only uses 1 thread).
  - exportWeightsAndBiases
```
public boolean exportWeightsAndBiases
```
    Whether to export Neural Network weights and biases to H2O Frames.
  - miniBatchSize
```
public int miniBatchSize
```
    Mini-batch size (smaller leads to better fit, larger can speed up and generalize better).
  - elasticAveraging
```
public boolean elasticAveraging
```
    Elastic averaging between compute nodes can improve distributed model convergence. #Experimental
  - elasticAveragingMovingRate
```
public double elasticAveragingMovingRate
```
    Elastic averaging moving rate (only if elastic averaging is enabled).
  - elasticAveragingRegularization
```
public double elasticAveragingRegularization
```
    Elastic averaging regularization strength (only if elastic averaging is enabled).
  - pretrainedAutoencoder
```
public ModelKeyV3 pretrainedAutoencoder
```
    Pretrained autoencoder model to initialize this model with.
- Constructor Detail
  - DeepLearningParametersV3
```
public DeepLearningParametersV3()
```
    Public constructor
- Method Detail
  - toString
```
public java.lang.String toString()
```
    Return the contents of this object as a JSON String.
    
    Overrides:
    
    toString in class ModelParametersSchemaV3

Class DeepLearningParametersV3

Field Summary

Fields inherited from class water.bindings.pojos.ModelParametersSchemaV3

Constructor Summary

Method Summary

Methods inherited from class java.lang.Object

Field Detail

balanceClasses

classSamplingFactors

maxAfterBalanceSize

maxConfusionMatrixSize

maxHitRatioK

activation

hidden

epochs

trainSamplesPerIteration

targetRatioCommToComp

seed

adaptiveRate

rho

epsilon

rate

rateAnnealing

rateDecay

momentumStart

momentumRamp

momentumStable

nesterovAcceleratedGradient

inputDropoutRatio

hiddenDropoutRatios

l1

l2

maxW2

initialWeightDistribution

initialWeightScale

initialWeights

initialBiases

loss

scoreInterval

scoreTrainingSamples

scoreValidationSamples

scoreDutyCycle

classificationStop

regressionStop

quietMode

scoreValidationSampling

overwriteWithBestModel

autoencoder

useAllFactorLevels

standardize

diagnostics

variableImportances

fastMode

forceLoadBalance

replicateTrainingData

singleNodeMode

shuffleTrainingData

missingValuesHandling

sparse

colMajor

averageActivation

sparsityBeta

maxCategoricalFeatures

reproducible

exportWeightsAndBiases

miniBatchSize

elasticAveraging

elasticAveragingMovingRate

elasticAveragingRegularization

pretrainedAutoencoder

Constructor Detail

DeepLearningParametersV3

Method Detail

toString