.. _parameters_H2ODeepLearning:
Parameters of H2ODeepLearning
-----------------------------
Affected Classes
################
- ``ai.h2o.sparkling.ml.algos.H2ODeepLearning``
- ``ai.h2o.sparkling.ml.algos.classification.H2ODeepLearningClassifier``
- ``ai.h2o.sparkling.ml.algos.regression.H2ODeepLearningRegressor``
Parameters
##########
- *Each parameter has also a corresponding getter and setter method.*
*(E.g.:* ``label`` *->* ``getLabel()`` *,* ``setLabel(...)`` *)*
activation
Activation function. Possible values are ``"Tanh"``, ``"TanhWithDropout"``, ``"Rectifier"``, ``"RectifierWithDropout"``, ``"Maxout"``, ``"MaxoutWithDropout"``, ``"ExpRectifier"``, ``"ExpRectifierWithDropout"``.
*Default value:* ``"Rectifier"``
*Also available on the trained model.*
adaptiveRate
Adaptive learning rate.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
ignoredCols
Names of columns to ignore for training.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
initialBiases
A array of weight vectors to be used for bias initialization of every network layer.If this parameter is set, the parameter 'initialWeights' has to be set as well.
*Scala default value:* ``null`` *; Python default value:* ``None``
initialWeights
A array of weight matrices to be used for initialization of the neural network. If this parameter is set, the parameter 'initialBiases' has to be set as well.
*Scala default value:* ``null`` *; Python default value:* ``None``
aucType
Set default multinomial AUC type. Possible values are ``"AUTO"``, ``"NONE"``, ``"MACRO_OVR"``, ``"WEIGHTED_OVR"``, ``"MACRO_OVO"``, ``"WEIGHTED_OVO"``.
*Default value:* ``"AUTO"``
*Also available on the trained model.*
averageActivation
Average activation for sparse auto-encoder. #Experimental.
*Default value:* ``0.0``
*Also available on the trained model.*
balanceClasses
Balance training data class counts via over/under-sampling (for imbalanced data).
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
calculateFeatureImportances
Compute variable importances for input features (Gedeon method) - can be slow for large networks.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
categoricalEncoding
Encoding scheme for categorical features. Possible values are ``"AUTO"``, ``"OneHotInternal"``, ``"OneHotExplicit"``, ``"Enum"``, ``"Binary"``, ``"Eigen"``, ``"LabelEncoder"``, ``"SortByResponse"``, ``"EnumLimited"``.
*Default value:* ``"AUTO"``
*Also available on the trained model.*
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.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
classificationStop
Stopping criterion for classification error fraction on training data (-1 to disable).
*Default value:* ``0.0``
*Also available on the trained model.*
columnsToCategorical
List of columns to convert to categorical before modelling
*Scala default value:* ``Array()`` *; Python default value:* ``[]``
convertInvalidNumbersToNa
If set to 'true', the model converts invalid numbers to NA during making predictions.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
convertUnknownCategoricalLevelsToNa
If set to 'true', the model converts unknown categorical levels to NA during making predictions.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
dataFrameSerializer
A full name of a serializer used for serialization and deserialization of Spark DataFrames to a JSON value within NullableDataFrameParam.
*Default value:* ``"ai.h2o.sparkling.utils.JSONDataFrameSerializer"``
*Also available on the trained model.*
detailedPredictionCol
Column containing additional prediction details, its content depends on the model type.
*Default value:* ``"detailed_prediction"``
*Also available on the trained model.*
diagnostics
Enable diagnostics for hidden layers.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
distribution
Distribution function. Possible values are ``"AUTO"``, ``"bernoulli"``, ``"quasibinomial"``, ``"modified_huber"``, ``"multinomial"``, ``"ordinal"``, ``"gaussian"``, ``"poisson"``, ``"gamma"``, ``"tweedie"``, ``"huber"``, ``"laplace"``, ``"quantile"``, ``"fractionalbinomial"``, ``"negativebinomial"``, ``"custom"``.
*Default value:* ``"AUTO"``
*Also available on the trained model.*
elasticAveraging
Elastic averaging between compute nodes can improve distributed model convergence. #Experimental.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
elasticAveragingMovingRate
Elastic averaging moving rate (only if elastic averaging is enabled).
*Default value:* ``0.9``
*Also available on the trained model.*
elasticAveragingRegularization
Elastic averaging regularization strength (only if elastic averaging is enabled).
*Default value:* ``0.001``
*Also available on the trained model.*
epochs
How many times the dataset should be iterated (streamed), can be fractional.
*Default value:* ``10.0``
*Also available on the trained model.*
epsilon
Adaptive learning rate smoothing factor (to avoid divisions by zero and allow progress).
*Scala default value:* ``1.0e-8`` *; Python default value:* ``1.0E-8``
*Also available on the trained model.*
exportCheckpointsDir
Automatically export generated models to this directory.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
exportWeightsAndBiases
Whether to export Neural Network weights and biases to H2O Frames.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
fastMode
Enable fast mode (minor approximation in back-propagation).
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
featuresCols
Name of feature columns
*Scala default value:* ``Array()`` *; Python default value:* ``[]``
*Also available on the trained model.*
foldAssignment
Cross-validation fold assignment scheme, if fold_column is not specified. The 'Stratified' option will stratify the folds based on the response variable, for classification problems. Possible values are ``"AUTO"``, ``"Random"``, ``"Modulo"``, ``"Stratified"``.
*Default value:* ``"AUTO"``
*Also available on the trained model.*
foldCol
Column with cross-validation fold index assignment per observation.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
forceLoadBalance
Force extra load balancing to increase training speed for small datasets (to keep all cores busy).
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
hidden
Hidden layer sizes (e.g. [100, 100]).
*Scala default value:* ``Array(200, 200)`` *; Python default value:* ``[200, 200]``
*Also available on the trained model.*
hiddenDropoutRatios
Hidden layer dropout ratios (can improve generalization), specify one value per hidden layer, defaults to 0.5.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
huberAlpha
Desired quantile for Huber/M-regression (threshold between quadratic and linear loss, must be between 0 and 1).
*Default value:* ``0.9``
*Also available on the trained model.*
ignoreConstCols
Ignore constant columns.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
initialWeightDistribution
Initial weight distribution. Possible values are ``"UniformAdaptive"``, ``"Uniform"``, ``"Normal"``.
*Default value:* ``"UniformAdaptive"``
*Also available on the trained model.*
initialWeightScale
Uniform: -value...value, Normal: stddev.
*Default value:* ``1.0``
*Also available on the trained model.*
inputDropoutRatio
Input layer dropout ratio (can improve generalization, try 0.1 or 0.2).
*Default value:* ``0.0``
*Also available on the trained model.*
keepBinaryModels
If set to true, all binary models created during execution of the ``fit`` method will be kept in DKV of H2O-3 cluster.
*Scala default value:* ``false`` *; Python default value:* ``False``
keepCrossValidationFoldAssignment
Whether to keep the cross-validation fold assignment.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
keepCrossValidationModels
Whether to keep the cross-validation models.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
keepCrossValidationPredictions
Whether to keep the predictions of the cross-validation models.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
l1
L1 regularization (can add stability and improve generalization, causes many weights to become 0).
*Default value:* ``0.0``
*Also available on the trained model.*
l2
L2 regularization (can add stability and improve generalization, causes many weights to be small.
*Default value:* ``0.0``
*Also available on the trained model.*
labelCol
Response variable column.
*Default value:* ``"label"``
*Also available on the trained model.*
loss
Loss function. Possible values are ``"Automatic"``, ``"Quadratic"``, ``"CrossEntropy"``, ``"ModifiedHuber"``, ``"Huber"``, ``"Absolute"``, ``"Quantile"``.
*Default value:* ``"Automatic"``
*Also available on the trained model.*
maxAfterBalanceSize
Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes.
*Scala default value:* ``5.0f`` *; Python default value:* ``5.0``
*Also available on the trained model.*
maxCategoricalFeatures
Max. number of categorical features, enforced via hashing. #Experimental.
*Default value:* ``2147483647``
*Also available on the trained model.*
maxRuntimeSecs
Maximum allowed runtime in seconds for model training. Use 0 to disable.
*Default value:* ``0.0``
*Also available on the trained model.*
maxW2
Constraint for squared sum of incoming weights per unit (e.g. for Rectifier).
*Scala default value:* ``3.402823e38f`` *; Python default value:* ``3.402823E38``
*Also available on the trained model.*
miniBatchSize
Mini-batch size (smaller leads to better fit, larger can speed up and generalize better).
*Default value:* ``1``
*Also available on the trained model.*
missingValuesHandling
Handling of missing values. Either MeanImputation or Skip. Possible values are ``"MeanImputation"``, ``"Skip"``.
*Default value:* ``"MeanImputation"``
*Also available on the trained model.*
modelId
Destination id for this model; auto-generated if not specified.
*Scala default value:* ``null`` *; Python default value:* ``None``
momentumRamp
Number of training samples for which momentum increases.
*Default value:* ``1000000.0``
*Also available on the trained model.*
momentumStable
Final momentum after the ramp is over (try 0.99).
*Default value:* ``0.0``
*Also available on the trained model.*
momentumStart
Initial momentum at the beginning of training (try 0.5).
*Default value:* ``0.0``
*Also available on the trained model.*
nesterovAcceleratedGradient
Use Nesterov accelerated gradient (recommended).
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
nfolds
Number of folds for K-fold cross-validation (0 to disable or >= 2).
*Default value:* ``0``
*Also available on the trained model.*
offsetCol
Offset column. This will be added to the combination of columns before applying the link function.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
overwriteWithBestModel
If enabled, override the final model with the best model found during training.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
predictionCol
Prediction column name
*Default value:* ``"prediction"``
*Also available on the trained model.*
quantileAlpha
Desired quantile for Quantile regression, must be between 0 and 1.
*Default value:* ``0.5``
*Also available on the trained model.*
quietMode
Enable quiet mode for less output to standard output.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
rate
Learning rate (higher => less stable, lower => slower convergence).
*Default value:* ``0.005``
*Also available on the trained model.*
rateAnnealing
Learning rate annealing: rate / (1 + rate_annealing * samples).
*Scala default value:* ``1.0e-6`` *; Python default value:* ``1.0E-6``
*Also available on the trained model.*
rateDecay
Learning rate decay factor between layers (N-th layer: rate * rate_decay ^ (n - 1).
*Default value:* ``1.0``
*Also available on the trained model.*
regressionStop
Stopping criterion for regression error (MSE) on training data (-1 to disable).
*Scala default value:* ``1.0e-6`` *; Python default value:* ``1.0E-6``
*Also available on the trained model.*
replicateTrainingData
Replicate the entire training dataset onto every node for faster training on small datasets.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
reproducible
Force reproducibility on small data (will be slow - only uses 1 thread).
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
rho
Adaptive learning rate time decay factor (similarity to prior updates).
*Default value:* ``0.99``
*Also available on the trained model.*
scoreDutyCycle
Maximum duty cycle fraction for scoring (lower: more training, higher: more scoring).
*Default value:* ``0.1``
*Also available on the trained model.*
scoreEachIteration
Whether to score during each iteration of model training.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
scoreInterval
Shortest time interval (in seconds) between model scoring.
*Default value:* ``5.0``
*Also available on the trained model.*
scoreTrainingSamples
Number of training set samples for scoring (0 for all).
*Scala default value:* ``10000L`` *; Python default value:* ``10000``
*Also available on the trained model.*
scoreValidationSamples
Number of validation set samples for scoring (0 for all).
*Scala default value:* ``0L`` *; Python default value:* ``0``
*Also available on the trained model.*
scoreValidationSampling
Method used to sample validation dataset for scoring. Possible values are ``"Uniform"``, ``"Stratified"``.
*Default value:* ``"Uniform"``
*Also available on the trained model.*
seed
Seed for random numbers (affects sampling) - Note: only reproducible when running single threaded.
*Scala default value:* ``-1L`` *; Python default value:* ``-1``
*Also available on the trained model.*
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).
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
singleNodeMode
Run on a single node for fine-tuning of model parameters.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
sparse
Sparse data handling (more efficient for data with lots of 0 values).
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
sparsityBeta
Sparsity regularization. #Experimental.
*Default value:* ``0.0``
*Also available on the trained model.*
splitRatio
Accepts values in range [0, 1.0] which determine how large part of dataset is used for training and for validation. For example, 0.8 -> 80% training 20% validation. This parameter is ignored when validationDataFrame is set.
*Default value:* ``1.0``
standardize
If enabled, automatically standardize the data. If disabled, the user must provide properly scaled input data.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
stoppingMetric
Metric to use for early stopping (AUTO: logloss for classification, deviance for regression and anomaly_score for Isolation Forest). Note that custom and custom_increasing can only be used in GBM and DRF with the Python client. Possible values are ``"AUTO"``, ``"deviance"``, ``"logloss"``, ``"MSE"``, ``"RMSE"``, ``"MAE"``, ``"RMSLE"``, ``"AUC"``, ``"AUCPR"``, ``"lift_top_group"``, ``"misclassification"``, ``"mean_per_class_error"``, ``"anomaly_score"``, ``"custom"``, ``"custom_increasing"``.
*Default value:* ``"AUTO"``
*Also available on the trained model.*
stoppingRounds
Early stopping based on convergence of stopping_metric. Stop if simple moving average of length k of the stopping_metric does not improve for k:=stopping_rounds scoring events (0 to disable).
*Default value:* ``5``
*Also available on the trained model.*
stoppingTolerance
Relative tolerance for metric-based stopping criterion (stop if relative improvement is not at least this much).
*Default value:* ``0.0``
*Also available on the trained model.*
targetRatioCommToComp
Target ratio of communication overhead to computation. Only for multi-node operation and train_samples_per_iteration = -2 (auto-tuning).
*Default value:* ``0.05``
*Also available on the trained model.*
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.
*Scala default value:* ``-2L`` *; Python default value:* ``-2``
*Also available on the trained model.*
tweediePower
Tweedie power for Tweedie regression, must be between 1 and 2.
*Default value:* ``1.5``
*Also available on the trained model.*
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.
*Scala default value:* ``true`` *; Python default value:* ``True``
*Also available on the trained model.*
validationDataFrame
A data frame dedicated for a validation of the trained model. If the parameters is not set,a validation frame created via the 'splitRatio' parameter. The parameter is not serializable!
*Scala default value:* ``null`` *; Python default value:* ``None``
weightCol
Column with observation weights. Giving some observation a weight of zero is equivalent to excluding it from the dataset; giving an observation a relative weight of 2 is equivalent to repeating that row twice. Negative weights are not allowed. Note: Weights are per-row observation weights and do not increase the size of the data frame. This is typically the number of times a row is repeated, but non-integer values are supported as well. During training, rows with higher weights matter more, due to the larger loss function pre-factor. If you set weight = 0 for a row, the returned prediction frame at that row is zero and this is incorrect. To get an accurate prediction, remove all rows with weight == 0.
*Scala default value:* ``null`` *; Python default value:* ``None``
*Also available on the trained model.*
withContributions
Enables or disables generating a sub-column of detailedPredictionCol containing Shapley values of original features.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
withLeafNodeAssignments
Enables or disables computation of leaf node assignments.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*
withStageResults
Enables or disables computation of stage results.
*Scala default value:* ``false`` *; Python default value:* ``False``
*Also available on the trained model.*