README.md

July 17, 2021 · View on GitHub

Nyata logo Logo by Assessoria de Comunicação do CeMEAI.

Nyata: visualization of Decision Tree and Random Forest models

  • Web platform to visualize and interpret Random Forests and Decision Tree models.
  • Compatible with scikit-learn tree and forest models.
  • Can be used with both Classification and Regression models.

Summary

This document is organized as follows:

  1. Installation instructions;
  2. Dependency list;
  3. Run instructions;
  4. Usage;
  5. Security considerations;
  6. Presenting Nyata main features.

Instalation instructions

First things first, you need to have Python 3, pip, node.js and npm, and angular 2 installed in your local machine in order to proceed with the Nyata installation.

After everything listed above checks out, now you need to install both the dependencies from the front-end and back-end. The script install.sh should do all the work for you with a bit of luck, if you run it giving your package manager as argument:

./install.sh <your package manager>

I wrote down the Makefile clauses for the following package managers: yay, pacman, apt, and apt-get. If your package manager is not in this list, just look into the backend/Makefile file and provide the installation commands by yourself.

./install.sh pacman

./install.sh yay

./install.sh apt

./install.sh apt-get

Dependency list

Nyata depends on the following third-party softwares:

  • Python 3
  • Packages for Python 3 (check backend/requirements.txt for the complete list)
  • Redis
  • npm
  • Angular 2
  • Tons of javascript/typescript/angular dependencies

Run instructions

After everything is installed correctly, you can run Nyata using run.sh:

./run.sh

If, for some reason, that does not work as expected, you can try to activate both back-end and front-end using its respectives Makefiles.

After activating the front-end, your default web browser should conveniently pop up with the Nyata application ready for use. If this is not the case, the front-end is served in http://localhost:4200 by default, and the back-end is served in http://localhost:5000.

Note that some web browsers does not support Nyata front-end.

Usage

After both backend and frontend are running, you can use Nyata directly from your web browser (http://localhost:4200). At first, you'll see an upload button asking for a .pickle file (Serialized Python Objects). The specific construction of this .pickle file for Nyata is a bit specific, so a function from the backend, nyata.dump, is available to create it for you after you give the required objects.

The visualization_example.ipynb jupyter notebook have some examples demonstrating the nyata.dump usage for Random Forests and Decision Trees.

Security considerations

If you're going to use the current version of Nyata in your system, remember that:

  • Do not use the public available secret key given in the backend/Makefile, since it is just for debugging purposes
  • Nyata receives an .pickle file as front-end input, and .pickle files are not secure. Hence, you'll probably want to keep Nyata as a personal tool, and not expose your system to the public.

Presenting Nyata main features

  • The platform works with sklearn Random Forest Classifiers and Regressors, and Decision Tree Classifiers and Regressors;
  • Model parameter summary and statistics; Main view of the platform.
  • Interactive view of every tree in the forest; View of an annotated tree of the forest model. Options available to show alongside the interactive view of the trees.
  • Prediction of a custom instance or an entire dataset of test instances; Highlighted prediction path of a custom test instance.
  • Analysis of the most common rules of the forest; and Analysis of the forest most common rules. Analysis of the forest most common rules, with feature names.
  • Tree hierarchical clustering based on prediction values or tree meta-characteristics. View of a hierarchical cluster performed on the forest trees.