HIPERMovelets

Source code of the paper HIPERMovelets: high-performance movelet extraction for trajectory classification, accepted for publication in International Journal of Geographical Information Science (2021).

[ publication ] [ preprint ] [ bibtex ]

Datasets

Datasets used in publication are available in 5-fold split in:

[ datasets ]

Versions

This is a project with the HIPERMovelets (Portela et al., 2021) implementation, with three options of optimizations.

HIPERMovelets: new optimization for MASTERMovelets, with greedy search (-version hiper).
HIPERMovelets-Log: plus, limits the movelets size to the ln size of the trajectory (-version hiper -Ms -3).
HIPERPivots: limits the movelets search space to the points that are neighbour of well qualified movelets of size one (-version hiper-pvt).
HIPERPivots-Log: plus, limits the movelets size to the ln size of the trajectory (-version hiper-pvt -Ms -3).

Includes re-implementations of the following methods (NOT EQUAL!):

MASTERMovelets: limits the movelets size to the ln size of the trajectory (-version 2.0).
MASTERMovelets-Log: limits the movelets size to the ln size of the trajectory (-version 2.0 -Ms -3).
MASTERMovelets-Pivots: it limit the movelets search space to the points that are neighbour of well qualified movelets of size one (-version pivots).
MASTERMovelets-Pivots-Log: it limit the movelets search space to the points that are neighbour of well qualified movelets of size one (-version pivots -Ms -3).

Setup

A. In order to run the code you first need to install Java 8 (or superior). Be sure to have enough RAM memory available.

(Optional)

B. If you opt to use the test automatization in Python (under dist/automatize folder), you first need to install Python, R, and dependencies.

See requirements.txt to Python dependencies. To install all dependencies you can use:

pip install -r ./dist/automatize/requirements.txt

Install R, and the dependent packages. To enter the R environment:

sudo apt-get install r-base
R

Install data.table and dplyr packages, and exit.

install.packages("data.table")
install.packages("dplyr")
q()

Usage

1. You can run the HIPERMovelets with the following command:

-curpath "$BASIC_PATH" 
-respath "$RESULT_PATH" 
-descfile "$DESC_FILE"  
-version hiper
-nt 8

Where:

BASIC_PATH: The path for the input CSV training and test files.
RESULT_PATH: The destination folder for CSV results files.
DESC_FILE: Path for the descriptor file. File that describes the dataset attributes and similarity measures.
-version: Method to run (hiper, hiper-pvt, ...)
-nt: Number of threads

2. For instance:

To run the HIPERMovelets you can run the java code with the following default entries as example:

java -Xmx80G -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper -nt 8 -ed true -samples 1 -sampleSize 0.5 -medium "none" -output "discrete" -lowm "false" -ms 1 -Ms -3 | tee -a "output.txt"

This will run with 80G memory limit, 8 threads, and save the output to the file output.txt`.

It is the same as (without the output file):

java -Xmx80G -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper -nt 8

Examples

HIPERMovelets

java -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper -nt 8

HIPERMovelets-Log

java -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper -nt 8 -Ms -3

HIPERMovelets-Pivots

java -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper-pvt -nt 8

HIPERMovelets-Pivots-Log

java -jar HIPERMovelets.jar 
-curpath "$BASIC_PATH" -respath "$RESULT_PATH" -descfile "$DESC_FILE" 
-version hiper-pvt -nt 8 -Ms -3

Extras

To run HIPERMovelts and classification methods with Python automatization (automatize) see the exemaples in the jupyter notebook: Automatize - Sample Code.ipynb

bigdata-ufsc/HiPerMovelets