A DFT-Based Running Time Prediction Algorithm for Web Queries

Oscar Rojas, Veronica Gil-Costa and Mauricio Marín
Explore the docs »

View train-test

Table of Contents

1. About The Project
   • run
2. Getting Started
   • Prerequisites
   • Installation
3. Usage
• main configuration
• data-web
4. Contributing
5. License

About The Project

Series of machine learning experiments and data mining tests for the evaluation of query execution time predictors in web search engines.

Techniques to predict the processing time of document ranking algorithms are compared, where predicting performance allows to allocate resources efficiently in web search engines.

run

• open file ./test.m and run with matlab

Getting Started

To get a local copy up and running follow these simple steps.

Prerequisites

matlab 2014 or higher

• https://la.mathworks.com/products/matlab.html

Installation

1. Clone the repo

   git clone https://github.com/neurovisionhub/dft-running-time-prediction.git

2. cd directory

   cd ./dft-running-time-prediction

Usage

In root folder, edit "test.m" to run predefined experiments ('simple','parallel','all','cross-fold','visual-map','block'), where 'configdefault.m' loads the experimental data.

clear
addpath(genpath('./'))
configdefault;
cd train-test\

% config examples
% Simple test Baseline vs DFT-Based 
% Selecction data-web
[data_simple,labelsD,labelsA]= main('simple');
data = [labelsD;labelsA;data_simple];
data = [["","", "Baseline","DFT-Based"]',data];
T = table(data)

% runtime-parallel prediction
%[data_parallel,~]= main('parallel');
.
.

main.m is a routine that receives as input a label (that describes the experiment) and outputs an array or tables.

Matrix Data

data_simple =

 0.9636    0.0219    0.8906    0.0356    0.9015    0.0490    0.9309    0.0330
 0.9688    0.0203    0.9279    0.0290    0.9235    0.0432    0.9324    0.0325

Table

                                    data                                    
 ___________________________________________________________________________

 ""             "W.CW.BM25"    "W.CW.BM25"    "BMW.CW.BM25"    "BMW.CW.BM25"
 ""             "PCr"          "RMSE"         "PCr"            "RMSE"       
 "Baseline"     "0.96363"      "0.021945"     "0.90145"        "0.049025"   
 "DFT-Based"    "0.96875"      "0.020327"     "0.92353"        "0.043161"

main-configuration

The training and test run configuration is set to ./train-test/main.m. General options include:

% ls = {0:local,1:shared}
% t = {0,1,2,3} = {1 thread, 2 threads, 4 threads, 8 threads}
% k = {10,100,1000,10000}  
% P = {0:baseline, 1:DFT-Based}
% p_test ={ %test }
%  sequence of neurons hidden layer for multiple test
vector_NHNeurons = [1,5,10,25,50];
% number of hidden neurons
NHNeurons = 5;

More information in train-test

data-web

Directory data-web contains:

• Queries Trec09 processed for ClueWeb and Gov2
• Descriptors of terms with baseline technique (for ClueWeb/Gov2 - BM25/TFIDF) and query 42 descriptors.
• Baseline technique query descriptors computed for k = 10,100,1000,10000 for ClueWeb with BM25

Contributing

Contributions are what make the open source community such an amazing place to be learn, inspire, and create. Any contributions you make are greatly appreciated.

License

Distributed under the MIT License. See LICENSE for more information.

Contact

O. Rojas - orojas.cl - [email protected]

Project Link: https://github.com/neurovisionhub/dft-running-time-prediction