Data Shaping Solutions offers software with source code
and technical support to perform various
statistical analyses. Our software is based on efficient, fast, robust
algorithms that can easily
handle large datasets. Thanks to creative statistical engineering, our
software solutions work well even with non Gaussian data, outliers
and ill-conditioned problems, where traditional packages would produce
an unstable solution. The source code is well documented and easy to understand or modify.
Our programs are written in platform-independent languages available at no cost
on any machine (Perl, C/C++, C#, SAS or Java). There is no licence fee.
We currently offer the following packages:
- Robust Multivariate Regression
Works well with non Gaussian data or outliers. Allows you to set up bounds
on the regression parameters (similar to ridge regression).
Does not use matrix inversion, thus
numerically stable. Robust parameter estimation based on Monte-Carlo
simulations and re-sampling. The source code can easily be modified to perform
This package can be used by scientists, programmers, analysts or engineers
with limited statistical knowledge.
Works on Unix, Linux or Windows.
We will help you install the software on your machine,
at no cost.
Price: Free. Click here to download.
- Offered with compact but simple and well documented source code (C, Perl or both).
- Processes datasets with hundreds of variables.
- Proprietary algorithm.
- Performs robust regression. Numerically stable.
- Performs thousands of regressions in a few seconds.
- Model validation.
- Confidence intervals and
percentiles for regression parameters using bootstrapping.
- Sensitivity analysis.
- Using different error minimization criteria.
- Confidence intervals for arbitrary combinations of regression parameters.
The first two modules are included in the package.
Example of possible use:
To perform a multiple regression where each regression coefficient has
the same sign as the correlation
between the dependent and associated independent variable. Such regressions are
more robust and more meaningful than traditional regressions: factors positively correlated
with the response always get a positive coefficient.