Statistica H2O Nodes

Sparkling Water is not shipped with Statistica and needs to be deployed, configured and started before Statistica H2O workspace nodes can be used.

H2O is an open source, in-memory, distributed, fast, and scalable machine learning and predictive analytics platform that allows the user to build machine learning models on big data. Sparkling Water allows users to combine the fast, scalable machine learning algorithms of H2O with the capabilities of Spark.

Statistica users may want to connect to H2O / Sparkling Water for two main reasons:

To use traditional algorithms and analyses in a distributed, scalable, in-memory environment for performance gains and practical feasibility of certain modeling designs, typically, involving large number of rows or columns in a dataset, which exceeds capabilities of a single host.
To have access to machine learning algorithms that are currently unavailable in Statistica .

H2O integration in Statistica can work with both H2O and Sparkling Water instances, deployed to clusters as well as single hosts. However, neither H2O nor Sparkling Water are shipped with Statistica and need to be deployed, configured and started before Statistica H2O workspace nodes can be used.

Please refer to the downloads page of H2O.ai website for installation instructions (https://www.h2o.ai/download/). We recommend using Sparkling Water 2.1 which runs with Spark 2.1.

Statistica nodes utilize H2O’s REST API, so H2O server needs to be accessible over the network. Typically, the H2O server IP address and port are presented to the user once H2O is started. This information is needed for the H2O nodes (H2O Data) to connect to the H2O environment.

NAVIGATION

The H2O nodes are found on the Big Data Analytics tab, when the workspace is in focus.

Importing Data into H2O

The process of data import in H2O involves two steps:

H2O Data Node

MISC

In this node you provide the H2O server IP address and port (mentioned in the previous section), and the path to the data source to be imported into H2O.

After running the node, it generates two outputs:

A H2O Inferred Variable Types report, which lists all variable names for the given data set and the data types inferred by H2O.
A spreadsheet that shows a sample of the dataset with the first few rows and inferred types.

H2O Data Mapping

1. Data Type Conversion

In this section, select which variables to use as Categorical or Continuous during the analysis.

If you do not select variables, the node will use the data types reported in the H2O Inferred Variable Types report of the H2O Data Node.

2. Data Frame

In this section, provide information for the following options:

Frame Name

In this field, enter the name for the new Frame to be created

Frame Type

In this field, select the name for the Frame Type (whether the data set will be used as Training or Validation data during the analysis)

Show first N cases

In this field, enter the number of cases to be displayed once the Frame is created.

H2O Data_Legacy Node [DEPRECATED]

You can still import Data into H2O with the H2O Data_Legacy node. However, be aware that this node always uses the Inferred data types and does not allow you to change the Variable data types. This node is deprecated and will not be available in future releases. You can import it by using the Node Browser. It is located in the DataMiner folder.

H2O Model Nodes

H2O Gradient Boosting Machine (GBM)

A GBM is an ensemble of either regression or classification tree models. Both are forward-learning ensemble methods that obtain predictive results using gradually improved estimations.

The H2O Gradient Boosting Machine(GBM) Node:

Attachments:
Required: One upstream node (H2O Data Mapping Node) with training data

Optional: A second upstream node (H2O Data Mapping Node) with validation data

It exposes H2O GBM model parameters organized in three sections:

Parameters

Advanced

Expert

In section four (Reports) select what outputs to generate. The output for the H2O Gradient Boosting Machine(GBM) Node includes the following reports:

Model parameters

Output (model category, validation metrics)

Model summary (number of trees, min. depth, max. depth, mean depth, min. leaves, max. leaves, mean leaves)

Scoring history in tabular format

Training metrics (model name, model checksum name, frame name, description, model category, duration in ms, scoring time, predictions, MSE, R2)

Variable importances in tabular format

H2O Generalized Linear Modeling (GLM)

Generalized Linear Models (GLMs) are an extension of traditional linear models. GLMs estimate regression models for outcomes following exponential distributions. They have gained popularity in statistical data analysis due to the following:

The flexibility of the model structure unifying the typical regression methods (such as linear regression and logistic regression for binary classification)

The recent availability of model-fitting software

The ability to scale well with large datasets

The GLM suite includes these analyses:

Gaussian regression

Poisson regression

Binomial regression (classification)

Multinomial classification

Gamma regression

The H2O Generalized Linear Modeling (GLM) Node:

Attachments:
Required: One upstream node (H2O Data Mapping Node) with training data

Optional: A second upstream node (H2O Data Mapping Node) with validation data

It exposes H2O GBM model parameters organized in three sections:

Parameters

Advanced

Expert

Reports, are where you can select what outputs to generate.   The output for the H2O Gradient Boosting Machine(GBM) Node includes the following reports:

Model parameters

Output (model category, validation metrics, and standardized coefficients magnitude)

GLM model summary (family, link, regularization, number of total predictors, number of active predictors, number of iterations, training frame)

Scoring history in tabular form (timestamp, duration, iteration, log likelihood, objective)

Training metrics (model, model checksum, frame, frame checksum, description, model category, scoring time, predictions, MSE, r2, residual deviance, null deviance, AIC, null degrees of freedom, residual degrees of freedom)

Coefficients

Standardized coefficient magnitudes (if standardization is enabled