This is mostly a continuation of the last post, but it's so nice!
As feedgnuplot reads data, it interprets it into separate datasets with IDs
that can be used to refer to these datasets. For instance you can pass
feedgnuplot --autolegend to create a legend for each dataset, labelling each
with its ID. Or you can set specific directives for one dataset but not another:
feedgnuplot --style position 'with lines' --y2 temperature would plot the
position data with lines, and the temperature data on the second y axis.
Let's say we were plotting a data stream
1 1 2 4 3 9 4 16 5 25
Without --domain this data would be interpreted like this:
- without
--dataid. This stream would be interpreted as two data sets: IDs0and1. There're 5 points in each one - with
--dataid. This stream would be interpreted as 5 different datasets with IDs1,2,3,4and5. Each of these datasets would contain point point each.
This is a silly example for --dataid, obviously. You'd instead have a dataset
like
temperature 34 position 4 temperature 35 position 5 temperature 36 position 6 temperature 37 position 7
and this would mean two datasets: temperature and position. This is nicely
flexible because it can be as sparse as you want: each row doesn't need to have
one temperature and one position, although in many datasets you would have
exactly this. Real datasets are often more complex:
1 temperature1 34 temperature2 35 position 4 2 temperature1 35 temperature2 36 3 temperature1 36 temperature2 33 4 temperature1 37 temperature2 32 position 7
Here the first column could be a domain of sort sort, time for instance. And we
have two different temperature sensors. And we don't always get a position
report for whatever reason. This works fine, but is verbose, and usually the
data is never stored in this way; I'd use awk to convert the data from its
original form into this form for plotting. Now that vnlog is a thing,
feedgnuplot has direct support for it, and this works like a 3rd way to get
dataset IDs: vnlog headers. I'd represent the above like this:
# time temperature1 temperature2 position 1 34 35 4 2 35 36 - 3 36 33 - 4 37 32 7
This would be the working representation; I'd log directly to this format, and work with this data even before plotting it. But I can now plot this directly:
$ < data.vnl
feedgnuplot --domain --vnlog --autolegend --with lines
--style position 'with points pt 7' --y2 position
I think the command above makes it clear what was intended. It looks like this:
The input data is now much more concise, I don't need a different format for plotting, and the amount of typing has been greatly reduced. And I can do the normal vnlog things. What if I want to plot only temperatures:
$ < data.vnl vnl-filter -p time,temp | feedgnuplot --domain --vnlog --autolegend --with lines
Nice!