Modularization

This release splits up the single java-timeseries package into two separate modules, math and timeseries. In addition, the arima package has undergone a major refactoring. Other notable changes.

• ARIMA models with drift. Drift terms are now available for modeling and forecasting. The semantics of the mean and drift terms align with those of the R forecast package.
• Time series linear regression models. Time series linear regression models may be created with relative ease. One feeds in a time series as a response variable and specifies whether to include an intercept, trend, and/or seasonal factors. Exogenous variables may also be added manually.
• Removal of unconditional-sum-of-squares (USS) fitting strategy. The strategies for fitting ARIMA models now coincide exactly with those in R. The default strategy is to use conditional-sum-of-squares (CSS) followed by maximum-likelihood (ML).

0.4 Pre-Release

This release includes a variety of enhancements to the library. The main changes are bugfixes, api improvements, more complete linear regression, and a new ARIMA simulation/process object.

Summary of major changes:

• Greatly enhanced linear regression functionality and API.
• ARIMA process object with potential for indefinite observation generation and memory-saving features.
• Package names now follow traditional Java conventions, preventing name clashes.
• Fix to prevent infinite looping in BFGS optimization.

0.2.1 Bugfix Release

• Replaces default compareTo implementation in FieldElement interface, which was only applicable to the Complex class, with class-specific implementations.
• Fixes build logic that exported slf4j implementation library.
• Broader test coverage.

Linear Regression

• Simple and multiple linear regression
• Various bugfixes
• Updated API documentation

Seasonal ARIMA

• Seasonal ARIMA modeling
• Basic time series analysis
• Simple plotting features.