Graphs and plots

What makes experiments most interesting is graphs. Or at least what they represent. To do this, my program generated a csv file with all the metrics it had gathered, and with that I loaded it into a second program that created the graphs. Here lay some mines to be aware of.

First of, how should the values be represented? If there are some deviates, most of the points may end up in a tiny cluster where we have no idea what they mean. If the points follow a general exponential curve, it might be better to represent the data logarithmic, meaning it might visually look linear, better enhancing the exponential point.

Second, it might be that the values have several decimals, meaning the bars might be cluttered with loads of numbers we generally don’t care about. The best way to solve this is to round it, even though it might look weird, we could fix that with some better indexing and labels.

Third, some graphs have some sort of relation, to better see similarities. One fix could be to put everything in one graph, but that could of course be too cluttered. Another could be to separate points in multiple graphs, just making sure they have the same limits to keep their relation. Just make sure that it is visually representing the data in a way that makes it clear what you want to get across.

Finally, only take the data that makes sense. If you got two types of data, the first being 10k iterations and the second being 1k, if they are identical, it’s better to pick the one that gives an even result. No need to show more than what is necessary.

On top of this, if you can explain the graphs, they may significantly improve the intention of the paper.