Our work on visual separation measures (EuroVis 2015, PacificVis 2016) relies on base data from our previous work (EuroVis 2012, InfoVis 2013).

In the paper "A Taxonomy of Visual Cluster Separation Factors" (EuroVis 2012), we created a set of 816 Scatterplots, both from synthetic and real datasets. Details about this base data can be found here.

For the paper "Empirical Guidance on Scatterplot and Dimension Reduction Technique Choices" (InfoVis 2013), two human coders manually judged the separability of 5460 classes in these scatterplots. we created a set of 816 Scatterplots, both from synthetic and real datasets. Details on these human judgements can be found here.

Data from EuroVis 2015 paper

For our work on the automatic evaluation framework (EuroVis 2015 paper), we thoroughly cleaned and aggregated the base data and human judgments from these two papers. This process resulted in 828 1-vs-all scatterplots (each with two classes), which are labeled either as "separable" or "non-separable" by a human viewer.

The following document gives a detailed description of the cleaning and aggregation process: Supplemental Material of EuroVis 2015 paper

Scatterplots

CSV Data: The following zip file contains the underlying data of these 828 two-class scatterplots:
→ 828 csv files (1.1 MB)

Images (PDF): The following zip file contains images of these 828 two-class scatterplots. The images were created with R for illustrative purposes only. These images were derived from the original multi-class scatterplots, on which the judgments were made (see Background and Base Data). The target class is shown in red, while the union of all other classes is shown in black:
→ 828 pdf images (5.3 MB)

Measure performance

15 state-of-the-art measures (CSV): Using these 828 scatterplots, we evaluated 15 state-of-the-art measures with different parameterizations. The following table lists all 828 scatterplots with measure performances:
→ csv (0.3 MB)
→ readme.txt with data description

Data and Code from PacificVis 2016 paper

Will be available soon...