ModellatoR

Description

Automates some parts of an analytics process, particularly the deployment of a model, its testing and reporting. It's built around the concept that people only have to do the preprocessing and specify 2 or 3 parameters. The rest is automatically done by our software.

It targets both power users and less knowledgable users. Power users can use directly the backend in an R console or through an IDE like Rstudio, and less knowledgable users can use the website.

It's licensed under an MIT License and freely available at github.

It can be used for the following analyses:

• Exploratory Analysis
  This describes a set of statistics and relationships among the input variables.
• Sensitivity Analysis
  This analyzes the contribution of the input variables towards an output variable.
• Performance Analysis
  This builds models and tests them. It can be used for Classification and Regression tasks using the following methods:
  • Decision Tree Models;
  • Random Forest Models;
  • Generalized Linear Models.

Revisions

Check NEWS.md for information about this.

Usage

Of the Backend

You can use it to create and setup projects, to train and test statistical models and also to generate automatic reports.

Typically you start by creating a folder structure using the function modellatoR::create_project and by doing its setup with the function modellatoR::setup_project. Now you have a folder structure similar to the one described in section Project's Folder Structure, and you can use this folder structure and its tools for your analytics process.

After these 2 steps you place the raw data in folder data and then if you load the project (ProjectTemplate::load.project()), the data will be imported automatically, and the scripts located in munge will be executed sequentially to preprocess and clean this data. Please note that either you need to feed clean raw data in folder data or you need to do these preprocessing scripts.

At this step the preprocessed resulting data structures can be optionally written into cache for future faster access. The preprocessed data is also loaded in the Environment and ready to be analysed through the scripts of src. The whole process is controlled by config/global.dcf and the reports are written into reports.

The model's parameters are defined in config/params.RData. If you need to change them, edit the list params accordingly and then save it into file config/params.RData.

If you need to change the report, adapt the corresponding Rmd file in reports.

Check section Project's Folder Structure for more information.

Of the GUI

In folder gui there is a Web Demo built using Shiny. To deploy it go to the root folder of the package and execute:

library(shiny)
runApp('gui')

Then your browser will open the modellatoR's GUI in a new tab. Fill in the sidebar with the requested information, click in Generate Reportand wait for the report to be ready. It will be downloaded automatically.

Project's Folder Structure

The folder structure is built on top of one created by ProjectTemplate and consists of the following:

• cache
  Here you'll store any data sets that are generated during a preprocessing step and don't need to be regenerated every single time you analyze your data. You can use the cache() function to store data to this directory automatically. Any data set found in both the cache and data directories will be drawn from cache instead of data based on ProjectTemplate's priority rules.

• config
  Here you'll store any configurations settings for your project.

• data
  Here you'll store your raw data files. If they are encoded in a supported file format, they'll automatically be loaded when you call load.project().

• gui
  Here are the files for the graphical interface.

• munge
  Here you can store any preprocessing or data munging code for your project. For example, if you need to add columns at runtime, merge normalized data sets or globally censor any data points, that code should be stored in the munge directory. The preprocessing scripts stored in munge will be executed sequentially when you call load.project(), so you should append numbers to the filenames to indicate their sequential order.

• reports
  Here are the report templates and the reports generated within the project.

• src
  Here you'll store your final statistical analysis scripts. You should add the following piece of code to the start of each analysis script: library('ProjectTemplate); load.project().

ProjectTemplate is intended to create statistical analysis projects and provides template code for data diagnostics, data munging, code profiling and unit testing. For its usage please check:

• Getting Started
• Mastering Project Template

Authors

This package was made by Antonio Rebordao.

I would like to acknowledge my employer Mentis SA for supporting this project.

License

This software is licensed under a MIT License. Check the file license.txt for more details.