Performance distribution of individual queries

This plot reports, as dots, the average performance in terms of queries per second against recall of several configruations of all the algorithms we testes, on all datasets.

The selection boxes allow to configure the combination of dataset, difficulty and algorithm.

Since the average performance obtained by each configuration of an algorithm is the result of several queries, we report, above and to the side of the plot, the distribution of the recall and query per seconds metrics for each individual query.

In some cases, we can observe a sort of bimodal behaviour: the average performance in terms of recall of some algorithms is given by a fraction of the queries having recall 0, while all the others have recall 1.

The running times, instead, are very concentrated around the average

Hovering with the mouse on a dot will stop the animation, focusing the visualization on that dot. Changing the configuration of the selectors will resume the animation