Huge plots number visualization/comparison

I am modelling different optimization methods for BER in MIMO system and i face difficulties when it comes to compare a new one algorithm with old previous ones(Lets say there're 20-30).

Ideally i would like to compare them using visualization methods.

Plotting all calculated BER is not easy to understand.

For now i have 1 temporary solution:

heatmap displaying BER text

Any ideas about better solutions?

Answer 1

[Clayton Gotberg]

You can't compare 20-30 algorithms visually without it being incredibly dense and difficult to understand. You could try:

1) Making several charts to compare a few old algorithms at a time to the new one

a) Maybe each chart contains a specific type of algorithm (I'm not familiar is but my guess is there are categories)

b) Maybe each chart is just the first three or five algorithms that haven't been compared already

2) Determine what you are using for comparison and chart that. For example, if you are comparing error over time, perhaps plot the average, maximum and minimum error for each algorithm instead.

3) Eliminate most of the algorithms before plotting. Perhaps plot only the best few and simply mention that you tested all of the others and that all resulted in worse performance than the ones in the plot. Maybe just pick a few that are most comparable to the new algorithm (like the ones that it's based on and the ones that it's most different from) to plot.

[Maxim Yuhlov]

Thank your for the ideas!

I tried categorizing plots, but there're still ~10 plots to compare.

Unfortunately, plot data might not be unique, but it's still unique algorithms and i can not ignore them. So there's no plot types for my situation?

[Clayton Gotberg]

Is this the kind of plot you're expecting to get back? I'm not an electrical engineer so I want to be sure I'm understanding the issue correctly.

If so, it looks like you can condense them into 3-4 plots in this way, but that is still a pretty large amount.

[Maxim Yuhlov]

Yes, you understood issue right. I plot constant system size(for example, 4x4), more snr values(so the plot is wider and the lines are drawn nearer to each other) and apply optimization algorithms. The number of plots depends on number of implemented optimizations(in some situation optimizations are combined).

[Clayton Gotberg]

You're definitely in a very tough spot because of the number of optimization techniques you've applied.

Since you have at most 30 algorithms, you can try grouping them in whatever way makes the most sense to you, with maybe 6 of them in each plot. That should keep them mostly readable but still makes 5 plots. If this is for an article, you could also just make a large number of plots but leave most of them in the supplementary material.

If you're really planning to reference all 30 of them in the body of the text, you will need plenty of plots to help break up and interpret your results - even just writing down 30 algorithms and a single observation about each (assuming they have names like in the above chart) is going to take up half a page with very dense text.