CHANGELOG.md
In robertzk/syberia: Syberia

Version 0.6.1.9008

Fixed test_engine to respect the global teardown hook.

Version 0.6.1.9007

Specify GitHub remotes in the DESCRIPTION file.

Version 0.6.1.9006

Fixed a negative test for test_engine.
Removed a memoization which keyed off a reference class object (and thus never actually memoized a value).

Version 0.6.1.9001-5

test_engine errors if there are any test failures.
Engine cloning uses HTTPS instead of SSH.
Mounting of engines are announced to the user.

Version 0.5.1-0.6.1

A complete re-write of the original Syberia that incorporates the new concept of engines, the core re-usable structural unit across different Syberia project (similar to Rails engines). All usages of syberia_project should be replaced with syberia_engine. The main bootstrapping logic that defines the convention in config/routes, lib and test have been moved to the base engine.

Minor bug fixes related to improvements in the base engine and modeling engine.

Version 0.4.3

A wide collection of legacy functions and helpers were removed. These should not affect any existing Syberia projects.

Version 0.4.2

Users can provide custom bootstrapping (i.e., things to do the first time the syberia project is loaded) by defining a config/boot resource. For example, placing cat("Bootstrapped") in config/boot.R will print "Bootstrapped" every time the syberia project is loaded for the first time in a fresh console.

Version 0.4.1

This version was released in conjunction with version 0.2.0 of syberiaStages and version 0.2.0 of syberiaStructure. The addition is the notion of Syberia IO "adapters".

IO adapters are (reference class) objects that have a \code{read} and \code{write} method. By wrapping things in an adapter, you do not have to worry about whether to use, e.g., \code{read.csv} versus \code{s3read} or \code{write.csv} versus \code{s3store}. If you are familiar with the tundra package, think of adapters as like tundra containers for importing and exporting data.

Version 0.4.0

syberia_version function for printing the current syberia version.
In data_stage, the mungebits are now called with train_only, which prevents them from getting trained so that the predict function does not accidentally run when re-running with the run helper method.
Customer tundra containers (i.e., classifiers) can now be stored in the lib/classifiers directory of your syberia project. If you do this, you need to use the same keyword as the first argument to the model_stage as the filename within the lib/classifiers directory, and include a function named train and predict.
When running a model, all of its helper files are also tracked for modifications, so that the cached environment for each step can be loaded. If this happens, Syberia will print "Copying cached environments..." when running the model (to prevent confusion when it doesn't happen -- for example, if you forgot to save one of the helpers).
Support for unit testing on individual model files. The unit tests must go in models/test under the same file path as your model version. So if you have a model in models/dev/models_one/model1.r, you must put unit tests in models/test/models_one/model1.r. This file must have a list structure similar to the one in the model, except the terminal nodes will be functions of the form function(before, after) { ... }, where before and after are environments holding the modeling environment before and after executing that step.
Better support for printing what models are running. The absolute file path will now be listed.