LASPATED: Library for Analysis of Spatio-Temporal Discrete Data

This repository contains the source code for the "LASPATED: A Library for the Analysis of Spatio-Temporal Discrete Data" paper (Arxiv link).

OS Support

LASPATED Python and MATLAB code were tested under Windows 10/Linux systems. The C++ code was tested under Linux systems only.

Installation

For installation instructions check INSTALL.md.

Usage

Python package

The following example shows basic usage of the python package:

app = spated.DataAggregator(crs="epsg:4326")

max_borders = gpd.read_file(
        r"../Data/rj/rj.shp"
    )  # Load the geometry of region of interest
app.add_max_borders(max_borders)

events = pd.read_csv(r"../Data/sorted_events.csv", encoding="ISO-8859-1", sep=",")
app.add_events_data(
    events,
    datetime_col="data_hora",
    lat_col="lat",
    lon_col="long",
    feature_cols=["prioridade"],
    datetime_format="%m/%d/%y %H:%M:%S",
)  # %m/%d/%y %H:%M:%S


app.add_time_discretization("m", 30, 60 * 24, column_name="hhs")

app.add_geo_discretization(
        discr_type="R", rect_discr_param_x=10, rect_discr_param_y=10
    )

app.plot_discretization()

First, the data aggregator is loaded and a Shapefile describing the city of Rio de Janeiro is used as a bounding border. Next, medical emergency data from Rio de Janeiro is loaded. Next, we define a time discretization of 30 minutes for each day and a space discretization of Rio de Janeiro into a grid with a maximum of 10 x 10 rectangles. Finally, we plot the space discretization, generating the following image:

C++ app

LASPATED also provides a C++ application, that given discretized data (possibly generated by the Python module), performs likelihood estimation of the intensities. After compiling the code (see the Installation guide), run:

./laspated --model_type=no_reg --method=calibration --algorithm=feasible
--info_file=info.txt --arrivals_file=arrivals.txt --neighbors_file=neighbors.txt --alpha_regions_file=alpha_regions.txt
--time_groups_file=time_groups.txt --durations_file=durations.txt
--output_file=output.txt

The command above will perform the likelihood estimation (method=calibration) using a projected gradient algorithm along the feasible region (algorithm=feasible) with a regularized model without covariates (model_type=no_reg). The user must also provide files with general data about the model (--info_file), data about the arrivals events (--arrivals_file), data about the regions under study (--neighbors_file), a space regularization matrix (--alpha_regions_file), a grouping of periods for time regularization (--time_groups_file) and a file describing the duration of each period. The estimated intensities are saved in the path provided by the --output_file option. For details about the format of each file, check the LASPATED user manual.

The user can also provide a configuration file and run the app with the -f option:

./laspated -f config.cfg

Data

Directory Data contains Shapefiles and data files that are used in the code examples. It includes a Rio de Janeiro and New York border Shapefiles, a csv and Shapefile with location of ambulance bases from Rio, covariates Data from Rio, and a Shapefile with Rio's administrative regions. It also includes csv files with Ambulance emergency data from Rio, and a mock csv simulating emergency data from New York.

laspated

Contains the laspated python module source code, that can be installed on your system via pip.

Model_Calibration

Contains the source code of the MATLAB, Octave and C++ functions.

Docker image

A docker container is available at DockerHub and also via the Dockerfile in the project root directory. See the INSTALL.md on how to generate the container.

vguigues/LASPATED