abeusher / petry-2020-marc

Source code for the paper "MARC: a robust method for multiple-aspect trajectory classification via space, time, and semantic embeddings"

MARC: a robust method for multiple-aspect trajectory classification via space, time, and semantic embeddings

Source code of the paper MARC: a robust method for multiple-aspect trajectory classification via space, time, and semantic embeddings, accepted for publication in International Journal of Geographical Information Science (IJGIS).

[ publication ] [ preprint ] [ bibtex ]

Setup

1. In order to run the code you first need to install all the Python dependencies listed in requirements.txt. You can do that with pip (works only with Python 3, tested with Python 3.6.7):

   pip install -r requirements.txt
   

2. Or if you use the Miniconda environment manager you can just run the following to create an environment with Python 3.6.7 and all required dependencies (replace ENV_NAME with whatever name you'd like):

   conda env create -f environment.yml --name ENV_NAME
   

   And then activate it with:

   conda activate ENV_NAME
   

Usage

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

   python multi_feature_classifier.py TRAIN_FILE TEST_FILE RESULTS_FILE DATASET_NAME EMBEDDING_SIZE MERGE_TYPE RNN_CELL
   

   Where:
   • TRAIN_FILE: The input CSV training file.
   • TEST_FILE: The input CSV test file.
   • RESULTS_FILE: The destination CSV results file.
   • DATASET_NAME: Label with the dataset name. This will be written in the results file as the dataset name.
   • EMBEDDING_SIZE: The embedding size of the attributes.
   • MERGE_TYPE: How attributes should be aggregated. It can be one of {add, average, concatenate}.
   • RNN_CELL: The recurrent cell used in the network. It can be one of {gru, lstm}.
2. For instance:

   python multi_feature_classifier.py data/foursquare_nyc/train.csv data/foursquare_nyc/test.csv results.csv FoursquareNYC 100 concatenate lstm