Ankur Goel

On hunt of awesomeness!

Bio-statsample-timeseries Deliverables

Hi there!

This summer, I’m working on new statsample-timeseries gem. This will act as an extension to existing Statsample by Claudio Bustos. I aim to add the support of timeseries and related functionalities to it.

I am using MiniTest for unit-testing and Cucumber for the feature testing. In the intial period of first phase, I targeted on completing as much portion of testing as possible for existing Statsample.

Also, another goal is to make statsample supported by current and previous Ruby versions. We are taking care of that - travis-ci.org/AnkurGel/statsample-timeseries by managing support of Ruby 2.0.0, 1.9.3, 1.9.2, jruby-19mode and rbx-19mode.

Now, we have many basic and advanced functions in place for timeseries. We have enabled:

  • Autocorrelation
  • Partial autocorrelation with:
    • yule-walker
    • levinson-durbin
    • biased levinson-durbin
  • Autocovariance
  • Lag and mean of series
  • correlation (almost complete)

Apart from them, we are also working on ARIMA module(another goal of project) and have realized the simulation of:

  • AR (Autoregressive) model
  • MA (Moving Average) model
  • ARMA model

For those simulations, the requisite was to pre-acquire the values of parameters against which the simulation was generated. For pure model, in this phase we aim to work and complete most of such functions to support that. We have completed:

  • Yule-walker for AR modelling.
  • Levinson-Durbin for AR modelling

I will now start with other such modelling like burg algorithm and IRLS. :).
The deliverables of project must be the all this stuff. Since the estimation methods for these modelling poses lot of theoretical and accuracy challenges, the pace to achieve them may not be too fast. :).

Documentation

One thing I liked in R and Statsmodels is amount of documentation and detail of API for developers and users. I wish to have similar amount of documentation for Statsample, so as to attract more number of Ruby developers and scientists.
Considering the amount of code present in Statsample and statsample-timeseries combined, maybe devoting considerable quality time for RDoc documentation will be a good idea!.

After

I already expressed my wish to continue contributing to project after timeline to Claudio. We will continue to work on it! On next module, if this is near completion. :)