davidealtomare/rego

Automatic Time Series Forecasting and Missing Values Imputation

*rego* is a machine learning algorithm for predicting and imputing time series. It can automatically set all the parameters needed, thus in the minimal configuration it only requires the target variable and the dependent variables if present. It can address large problems with hundreds or thousands of dependent variables and problems in which the number of dependent variables is greater than the number of observations. Moreover it can be used not only for time series but also for any other real valued target variable. The algorithm implemented includes a Bayesian stochastic search methodology for model selection and a robust estimation based on bootstrapping. *rego* is fast because all the code is C++.

Installation from PyPi

pip install --upgrade setuptools
pip install wheel
pip install Cython
pip install pandas
pip install rego

Installation from CRAN

R --vanilla -e 'install.packages(c("Rcpp", "RcppArmadillo"), repos="http://cran.us.r-project.org")'
R --vanilla -e 'install.packages(c("rego"), repos="http://cran.us.r-project.org")'

Compilation notes

Compilation requires a C++11 compiler. For a debian based SO, they can be easily installed running:

apt-get install build-essential

Compilation on Windows requires Microsoft Visual C++ 14.0 or greater. (https://visualstudio.microsoft.com/it/downloads/)

Compiling and testing C++ code

sh compile/compile-cpp.sh test/test.cpp
/tmp/build/c++/main.o

Compiling and testing Python code

Note! Only Python3 is supported!

pip install -r python/requirements.txt


sh compile/compile-py.sh
python test/test.py 

Compiling and testing R code

sh compile/compile-R.sh
Rscript test/test.R

Generating Python documentation

pip install sphinx
pip install numpydoc
pip install pygments --upgrade
pip install rinotype

cd python/src/cypack

python generate_doc.py

The following .pdf will be generated:

python/src/cypack/docs/_build/rinoh/rego.pdf