TrajPop is a tool that has been developed in the ERC GeoDiverCity project.
It allows user to create clusters of cities according to their temporal population evolution, resulting in clusters of (temporal) Trajectories of Populations.
For a brief demo of the application possibilities, click on the
load test datasetbutton
This application expects a plain text document as input (ie. a csv or tsv file), following this data formatting as minimal requirement:
Note that it can also contains other variables, such as lat/long data for mapping, or categorical data for correspondence analysis.
Depending on your text format, you might have to check
CSV Import Settings, which will allow you to configure the application accordingly to your data.
Once a file is loaded, there's 2 needs :
ID column, that needs to be a unique attribute of each feature.
Temporal columns(3 minimum) that will be used to define the clusters.
The plots are computed accordingly to the number of clusters defined with the slider
Number of wanted clusters. The colors are defined using the
Color palette selector, and refers to RColorBrewer qualitative palettes. Note that those palettes can only be used for a defined number of colors, so, when trying to draw more colors than the palette allows, the palette choice will be switched to a rainbow palette.
The Table panel contains a dynamic version of the originally uploaded table, with the addition of the cluster each feature belongs to. The
Search function might prove really useful when trying to identify the cluster a city is in. Note that using the
Download Table button, you can export this increased table, which you'll download as a CSV file.
Please keep in mind that heavy tables can take some time to load, so, don't hesitate looking at other tabs while it's loading.
This tab features two maps of your data, as long as you fill the
Longitude inputs, and also the
Scales points on one. This last argument must be applied on a numeric column, and all the points size will then be computed out of the maximum value of this column, having a maximum radius that can be customized with the
Max. point size slider. Note that when creating the pdf report, this value will also be used for the static map, so, you might want to configure it for this static output when you'll need to build the report.
This tab allows you to perform correspondence analysis on the computed clusters. This means that you can compare the elements composing those clusters to various variables of yours, depending on their
Log10box. It sets the Y axis to a log scale. Beside the box-plot, a Fisher-Snedecor (F-test) test is run on the variables, determining also the independence of the variable.
TrajPop performs a Correspondence Analysis on a temporal population table. The coordinates of cities on the orthogonal components then make it possible, after re-entering the weights of the features, to generate a matrix for population discrepancies amond cities (measured using a chi² distance). To this matrix is applied a Hierarchical Cluster Analysis (using the Ward method, which tends to minimise intra-class variance and to maximise inter-class variance).
From the tree generated by this clustering, the user have to choose a number of clusters that sufficiently distinguishes the trajectories while at the same time enables them to be mapped. It is then possible to analyse trajectory classes, using the various graphical outputs such as plots and maps.
Here are the most common (and easily correctible) sources of errors.
We'd like to thanks the creator of those packages, and especially the maintainers of Shiny, for their valuables advices and answers.
Reference : Robin Cura, (2013). TrajPop (1.0 web version) [Web application]. Retrieved from http://trajpop.parisgeo.cnrs.fr
TrajPop v1.0.1 - October 2013
Copyright 2013 Robin Cura, Géographie-cités