Introduction

Welcome to the QuickRT home page. QuickRT is a highly optimized Windows command-line program that implements an Extremely Randomized Trees (ExtraTrees) classifier. The Extremely Randomized Trees classifier is a derivative of the Random Forest classifier by Leo Breiman (see the Wikipedia Random Forest article). QuickRT is highly optimized for both speed and memory, is multi-threaded, and takes advantage of SIMD instructions of modern processors.

The main benefit of QuickRT is its speed and low memory usage. For classification on larger data sets, speed translates into accuracy. The faster the implementation the more trees and more thorough node split searches you can do. Depending on various factors, more trees and more thorough searches increase the accuracy of the classifier.

The Preliminary Performance Analysis page has information and benchmarks comparing QuickRT to other classifier implementations.

The current version is the first alpha test version (0.0.0). See the Download section below to download the installer. Please send feedback and questions to ryan@TetrachromeSoftware.com

Use

QuickRT uses csv (comma-separated values) files. A typical csv file has a header line that specifies the column names, and then one line per sample. QuickRT requires an integer column that identifies the sample (often called "index" or "id"), one or more feature columns (floating point or integer), and a "class" column that specifies the true classification of each sample. (Note that the "index" and "class" column names can be specified via arguments.)

The following are the main modes of operation:

  • Train the classifier from a csv file (--input argument) and perform prediction on samples from another csv file (--test argument).
  • Perform K-fold validation on samples from a csv file (--input argument).
In both cases the per-sample results can be written to a result csv file.

You can see the full help here.

Features

The following are primary features of QuickRT:

  • Highly optimized for speed and memory use.
  • 64-bit so it supports large data sets.
  • Multi-threaded for efficient use of CPU resources.
  • Performs binary (two-class) and multi-class classification. (Regression is not supported yet.)
  • Performs K-fold validation on samples from a csv file.
  • Arguments to select which columns to use.
  • Arguments to specify classifier parameters such as number of trees, speed-vs-accuracy, max tree depth, node split metric, and target node size.
  • Arguments to specify random number generator seed and the number of repetitions to run. When multiple repetitions are run the results are averaged on a per-sample basis.
  • Performs logging to a text file and has arguments to control verbosity.
  • Arguments to specify how many folds for K-fold validation, or to do random vs blocked folding, or spread-and-striped folding.

Download

See the EULA that is included in the installer for the official disclaimer of warranty.

Version Release Date Link MD5
0.0.0 May 27, 2015 QuickRT_0.0.0.msi 41E4AF5AC26005AF557E01214F0FC473

Currently Windows 8 SmartScreen blocks the QuickRT installation due to the fact that not many people have installed QuickRT yet. Windows 8 shows a SmartScreen dialog that indicates that the installation has been blocked, and unfortunately by default it does not show you the option to proceed with the installation. You have to select the details button to see the option to allow the installation to continue. There is no malware in this installer and you can check it with various web-based virus scanners like the one linked below.

As a new program, there is the possibility that virus detection strategies (in addition to SmartScreen) that rely on a history of successful uses may be triggered, despite there being no malware in QuickRT. There are free services such as VirSCAN.org that you can use to scan files for malware.