small Rosettas for Python, scikit-learn, R, Flow, Sparkling Water for the website

[exalate-issue-sync[bot]]

Tiny examples in each language for each algo, for the website.

[exalate-issue-sync[bot]]

Neeraja Madabhushi commented: We need list of missing Rosettas for all clients.

[exalate-issue-sync[bot]]

J commented: More info for when there is time to work on this:

• should be demos
• include primary functions (import, parse, build model, etc)
• look at booklets
• should appear as tabs for each lang. (like tabs on downloads page)

[exalate-issue-sync[bot]]

J commented: Preliminary list of languages:

• R
• Python
• scikit-learn
• Flow (I can help out here)
• Sparkling Water
• others??

Preliminary list of functions for Rosettas:

• install H2O
• point to another H2O instance (connect to a cluster)
  • start H2O
• shutdown H2O
• build model
• save model
• import model
• export model
• import files
• upload files
• get frames
• split frame
• get models
• get predictions
• get jobs
• predict
• view logs
• download logs
• create synthetic frame
• get cluster status
• view all frames
• impute

[~accountid:557058:3ae3c86a-e56a-4211-99d4-9a8cf5ab63f6], can you please review and update as needed? I'm sure there are other functions that should be included here, but I thought this would be a good starting point. Thanks!

[exalate-issue-sync[bot]]

Raymond Peck commented: The overarching goal is for this to be:

• a quick start, so people who work best by following examples can get going really fast, and
• an illustration of / reference for the differences between the language bindings.

It should also link to the reference docs, where appropriate.

For "build model":

• there should be examples for each algo, and for each model category (regression, binomial classification, multinomial classification, clustering, autoencoder) that each can do.
• the fact that regression v. classification is chosen by the type of the response column should be made very clear; e.g., examples should show as.factor to trigger classification. If we have a good example, doing regression and classification on the same dataset using the same response would make this very clear.
• use of validation frames with test/train split v. using cross-validation
• looking at model metrics, including plots
• use of model metrics to compare performance of two models

Add grid search.

Add data conversion (R data frame and data.table <-> H2O, Python Pandas / numpy / raw 2d arrays <-> H2O, RDD <-> H2O).

Import Files needs to include single file case + directory case.

Split frame should have both random (seeded) runif, plus row slice.

Add row and column slicing by name and index.

User row weights example.

Add examples for handling of unbalanced datasets both using Balance Classes and using user row weights.

Add examples of using GLRM and PCA to reduce dimensionality to feed into other model builders.

[DinukaH2O]

JIRA Issue Migration Info

Jira Issue: PUBDEV-1898 Assignee: Joby Joy Reporter: Raymond Peck State: Open Fix Version: N/A Attachments: N/A Development PRs: N/A