H2O Neural Network

H2O Deep Learning is based on a multi-layer feedforward artificial neural network that is trained with stochastic gradient descent using back-propagation.

Input

It takes in a DataFrame as input

Type

ml-estimator

Class

fire.nodes.h2o.NodeH2ONeuralNetwork

Fields

Name	Title	Description
isResponseIsCategorical	isResponseColIsCategorical	Specify a response column type(numeric or categorical). Separates the Classification and Regression
labelCol	Label Column	Response variable column.
ignoredCols	Ignored Columns	Features to be ignored for Modelling
splitRatio	Split Ratio	Split Ratio
columnsToCategorical	Columns to Categorical	Columns to be Categorical encoded
seed	Seed	Seed for pseudo random number generator (if applicable).
balanceClasses	Balance Classes	Balance training data class counts via over/under-sampling (for imbalanced data).
maxAfterBalanceSize	Max After Balance Size	Maximum relative size of the training data after balancing class counts (can be less than 1.0). Requires balance_classes.
activation	Activation	Activation function.
hidden	Hidden	Specify the hidden layer sizes (value must be positive)
epochs	Epochs	SHow many times the dataset should be iterated (streamed), can be fractional.
nfolds	Number of Folds	Number of folds for K-fold cross-validation (0 to disable or >= 2).
trainSamplesPerIteration	Train Samples Per Iteration	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	Target ratio comm to comp	Target ratio of communication overhead to computation. Only for multi-node operation and train_samples_per_iteration = -2 (auto-tuning).
categoricalEncoding	Categorical Encoding	Specify one of the various encoding schemes for handling categorical features
ignoreConstCols	Ignore Const Columns	Ignore constant columns.
scoreEachIteration	Score Each Iteration	Whether to score during each iteration of model training.
stoppingRounds	Stopping Rounds	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).
maxRuntimeSecs	Max Runtime Secs	his argument specifies the maximum time that the AutoML process will run for. If both max_runtime_secs and max_models are specified, then the AutoML run will stop as soon as it hits either of these limits. If neither max_runtime_secs nor max_models are specified, then max_runtime_secs defaults to 3600 seconds (1 hour).
stoppingMetric	StoppingMetric	Metric to use for early stopping (AUTO: logloss for classification, deviance for regression)
stoppingTolerance	StoppingTolerance	Relative tolerance for metric-based stopping criterion (stop if relative improvement is not at least this much)
standardize	Standardize	If enabled, automatically standardize the data. If disabled, the user must provide properly scaled input data.
loss	Loss	SLoss function.
adaptiveRate	Adaptive Rate	Adaptive learning rate.
rho	Rho	Adaptive learning rate time decay factor (similarity to prior updates).
advanced	Advanced
epsilon	Epsilon	Adaptive learning rate smoothing factor (to avoid divisions by zero and allow progress).
rate	Rate	Learning rate (higher => less stable, lower => slower convergence).
rateAnnealing	Rate Annealing	Learning rate annealing: rate / (1 + rate_annealing * samples).
rateDecay	Rate Decay	Learning rate decay factor between layers (N-th layer: rate * rate_decay ^ (n - 1)
momentumStart	Momentum Start	Specify the initial momentum at the beginning of training; we suggest 0.5 (Applicable only if adaptive_rate is disabled)
momentumRamp	Momentum Ramp	Number of training samples for which momentum increases.
momentumStable	Momentum Stable	Final momentum after the ramp is over (try 0.99).
nesterovAcceleratedGradient	Nesterov Accelerated Gradient	Use Nesterov accelerated gradient (recommended).
inputDropoutRatio	Input Dropout Ratio	Input layer dropout ratio (can improve generalization, try 0.1 or 0.2).
inputDropoutRatio	Hidden Dropout Ratios	Hidden layer dropout ratios (can improve generalization), specify one value per hidden layer, defaults to 0.5.
l1	L1	L1 regularization (can add stability and improve generalization, causes many weights to become 0).
l2	L2	L2 regularization (can add stability and improve generalization, causes many weights to be small.
maxW2	Max W2	Constraint for squared sum of incoming weights per unit (e.g. for Rectifier).
initialWeightDistribution	Initial Weight Distribution	Initial weight distribution.
initialWeightScale	Initial Weight Scale	Uniform: -value…value, Normal: stddev.
scoreInterval	Score interval	Shortest time interval (in seconds) between model scoring.
scoreTrainingSamples	Score Training Samples	Number of training set samples for scoring (0 for all).
scoreValidationSamples	Score Validation Samples	Number of validation set samples for scoring (0 for all).
scoreDutyCycle	Score Duty Cycle	Maximum duty cycle fraction for scoring (lower: more training, higher: more scoring).
classificationStop	Classification Stop	Stopping criterion for classification error fraction on training data (-1 to disable).
regressionStop	Regression Stop	Stopping criterion for regression error (MSE) on training data (-1 to disable).
quietMode	Quiet mode	Enable quiet mode for less output to standard output.
scoreValidationSampling	Score Validation Sampling	Method used to sample validation dataset for scoring.
overwriteWithBestModel	Overwrite With Best Model	If enabled, override the final model with the best model found during training.
useAllFactorLevels	Use All Factor Levels	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.
diagnostics	Diagnostics	Enable diagnostics for hidden layers.
calculateFeatureImportances	Calculate Feature Importances	Compute variable importances for input features (Gedeon method) - can be slow for large networks.
fastMode	Fast Mode	SEnable fast mode (minor approximation in back-propagation).
forceLoadBalance	Force Load Balance	Force extra load balancing to increase training speed for small datasets (to keep all cores busy).
replicateTrainingData	Replicate Training Data	Replicate the entire training dataset onto every node for faster training on small datasets.
singleNodeMode	Single Node Mode	Run on a single node for fine-tuning of model parameters.
shuffleTrainingData	Shuffle Training Data	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	Missing Values Handling	Handling of missing values. Either MeanImputation or Skip.
sparse	Sparse	Sparse data handling (more efficient for data with lots of 0 values).
averageActivation	Average Activation	Average activation for sparse auto-encoder. #Experimental.
sparsityBeta	Sparsity Beta	Sparsity regularization. #Experimental.
maxCategoricalFeatures	Max Categorical Features	Max. number of categorical features, enforced via hashing. #Experimental.
reproducible	Reproducible	Force reproducibility on small data (will be slow - only uses 1 thread).
exportWeightsAndBiases	Export Weights And Biases	Whether to export Neural Network weights and biases to H2O Frames.
miniBatchSize	Mini Batch Size	SMini-batch size (smaller leads to better fit, larger can speed up and generalize better).
elasticAveraging	Elastic Averaging	Elastic averaging between compute nodes can improve distributed model convergence. #Experimental.
elasticAveragingMovingRate	Elastic Averaging Moving Rate	Elastic averaging moving rate (only if elastic averaging is enabled).
elasticAveragingRegularization	Elastic Averaging Regularization	Elastic averaging regularization strength (only if elastic averaging is enabled).
keepCrossValidationModels	Keep Cross Validation Models	Whether to keep the cross-validated models. Keeping cross-validation models may consume significantly more memory in the H2O cluster.
keepCrossValidationPredictions	Keep Cross Validation Predictions	Whether to keep the predictions of the cross-validation predictions. This needs to be set to TRUE if running the same AutoML object for repeated runs because CV predictions are required to build additional Stacked Ensemble models in AutoML.
keepCrossValidationFoldAssignment	Keep Cross Validation Fold Assignment	Whether to keep cross-validation assignments.
distribution	Distribution	Distribution function.)
tweediePower	Tweedie Power	Tweedie power for Tweedie regression, must be between 1 and 2.
quantileAlpha	Quantile Alhpa	Desired quantile for Quantile regression, must be between 0 and 1.
huberAlpha	Huber Alpha	Desired quantile for Huber/M-regression (threshold between quadratic and linear loss, must be between 0 and 1).
weightCol	Weight Column	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.
offsetCol	Offset Column	Offset column. This will be added to the combination of columns before applying the link function.
foldCol	Fold Column	Column with cross-validation fold index assignment per observation.
foldAssignment	Fold Assignment	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.
aucType	AUC Type	Set default multinomial AUC type.
confusionMatrix	Confusion Matrix
output_confusion_matrix_chart	Output Confusion Matrix Chart	whether to display confusion matrix chart.
cm_chart_title	Confusion Matrix Chart Title	Title name to display in Confusion Matrix Chart
cm_chart_description	Confusion Matrix Chart Description	Description to display in Confusion Matrix CHart
confusionMatrixTargetLegend	Confusion Matrix Target Legend	Legend name to display for Target in Confusion Matrix
confusionMatrixPredictedLabelLegend	Confusion Matrix PredictedLabel Legend	Legend name to display for Predicted Label in Confusion Matrix
ROC Curve	ROC Curve
output_roc_curve	Output ROC Curve	Whether to display confusion matrix chart.
roc_title	ROC Curve Chart Title	Title name to display in ROC Curve Chart
roc_description	ROC Curve Chart Description	Add Description for ROC Curve Chart
xlabel	X Label	X label
ylabel	Y Label	Y Label
Grid Search	Grid Search
paramKeys	Param Name	Param Names. eg: l1 ,hidden
paramValues	Param Value	Enter comma separated values.eg: 0, 1e-5, eg: 50,50;100,100
gridStrategy	Grid Search Strategy	Strategy to use for model hyperparameter search. Cartesian does exhaustive search; RandomDiscrete searches randomly within given time or model limits.
gridMaxModels	Grid Max Models	Maximum number of models to build in the grid search (0 for unlimited).
gridMaxRuntimeSecs	Grid Max Runtime Seconds	Maximum runtime in seconds for the grid search (0 for unlimited).
gridStoppingRounds	Grid Stopping Rounds	Early stopping based on convergence of the metric during grid search (0 to disable).
gridStoppingTolerance	Grid Stopping Tolerance	Tolerance for metric-based stopping criterion during grid search.
gridStoppingMetric	Grid Stopping Metric	Metric to use for early stopping during grid search (AUTO: logloss for classification, deviance for regression).
gridParallelism	Grid Parallelism	Level of parallelism to use when building models in the grid.
gridSelectBestModelBy	Grid Select Best Model By	Metric used to select the best model from the grid.

Details

H2O’s Deep Learning is based on a multi-layer feedforward artificial neural network that is trained with stochastic gradient descent using back-propagation. The network can contain a large number of hidden layers consisting of neurons with tanh, rectifier, and maxout activation functions.

More details are available at : http://docs.h2o.ai/h2o/latest-stable/h2o-docs/data-science/deep-learning.html