Dynamic navigation does not work for a single algorithm

[oskargirardin]

When post-processing an experiment with a single algorithm, the dynamic navigation feature does not show the ECDF plots (despite the link being shown). Here is a screenshot of the page.

[nikohansen]

As a quick-ish fix, I'd suggest to change the processing logic in that pprldmany becomes the default way even for a single algorithm and --single-algorithm becomes an option for invoking the original processing.

To accomplish this relatively easily, we can add random search to the algorithms list if the list contains only one element and not --single-algorithm. When catching random search fails, we should give a hint to use the --single-algorithm option.

[brockho]

Just for information: for all of bbob, bbob-biobj, bbob-boxed, bbob-constrained, bbob-largescale, bbob-mixint, and bbob-noisy, we already have a random search entry in the data archive.

[nikohansen]

Are they all under the same name?

[brockho]

No, because we have for some test suites several versions of random search. But for all test suites, using the first of the random search entries via randomsearch! should do. Only for bbob-biobj, it will pick the random search, sampling in $[-4,4]^n$, for all other test suites, the first entry is one that sampled in $[-5,5]^n$.

[nikohansen]

In order to get the right randomsearch, we need to first identify the appropriate "current testbed" and get then the data set run on this testbed/suite. That is, unfortunately, we can not append just randomsearch! to the argument string.

[nikohansen]

It should also be relatively simple to

cp pprldflex.html pprldflex1.html

and change the link entries to the correct names. Sweeping through dimension is not necessary or possible anymore. The entries "per function" and "single plot" are the same, as all dimensions are in the same plot.