startml version 0.1.3

Added deep learning grid search to, feature and weak learner prediction mixing, and PCA for weak learner predictions to ensemble function.

First fully functional workflow on regression models:
startml ->plot -> trim -> plot -> ensemble -> plot -> plot_dlayer

In just a few commands, hundreds of models can be build, inspected or compared, selected, a hyper parameter ensemble search, visually compare ensemble and weak learners, and visualize the layers of of the deep learning ensemble.

Running startml for 40 minutes total building deep learning and gradient boosted machines (20 minutes each) feed into an ensemble with the deep learning algorithm grid search for 200 seconds, keep_features = TRUE, and PCA dimension reduction of 70% scores in top 25 percent of kaggle Ames housing prices "for fun" contest.

startml version 0.1.2

The plot function has been updated to include test performance of models, and plot an ensemble if it exists in mlblob.

Additionally a function was added called trim which can select model based on correlation and/or performance thresholds after a mlblob is created. This allows for a iterative and more flexible workflow.

An ensemble function was added which creates an ensemble model from an mlblob object. Currently only mean of weak learners is implemented. Planned options include deeplearning, tree methods for meta learners, dimensional reduction, and feature / weak learner predictions mixing.

startml version 0.1.1

Some errors were corrected in namespace.

startml Version 0.1.0

Installable R package version of startml.

Notes:

• Installed and ran successfully on windows 7, windows 10, and OSX 10.10 with Rstudio Version 1.0.143 and R-3.4.0 Patched x64
• Extras in this version that may be useful as standalone functions when using H2O include the plot_dlayer function, a way to visualize hidden layers of H20 deep learning models in 2 or 3 dimensions.
• Installs with devtools function install_github