Statement of purpose: this room is for discussions about a new HEP module in Matplotlib, as a centralized/official way of making common ROOT plots in Matplotlib. It should collect many scripts and libraries that physicists have written to do this through the current interface (sometimes by hacking the current interface).
Some of the important issues include
- plotting pre-histogrammed data, supplying only the bin contents;
- handling log scales gracefully, considering that some bins may be 0 (unfilled) or negative (due to the negative weights we frequently use, a bin can fluctuate down);
- the "skyline" visual rendering that's conventional in our field;
- error bars and details of the error bar rendering (e.g. do we need caps at the ends?);
- common plot types in our field: stacked simulations, data with error bars, difference between the stacked simulation and the data in an aligned window, overlaid fits, confidence bands, upper limit plots, etc.;
- common and Pythonic data formats to deliver such data to Matplotlib: e.g. do we want to extend Numpy's 2-tuple (bin contents and axis edges) to include more information in subsequent tuple indexes? If so, can we standardize a format?
rootpy (http://www.rootpy.org/auto_examples/plotting/plot_matplotlib_hist.html) is one example of a library that makes Matplotlib behave as we want; link others here.
Note the drop toward zero at the left and right edges and between zero and non-zero bins, even in log scale.
Low statistics: 
Reminder of the google doc: https://docs.google.com/document/d/1ju3JObrPEyT7QpCx3m1YoVp0gFbLay8ZIxtO0mUCvgo/edit#heading=h.3tw53mc90yd
The google doc is excellent, I forgot we had that
I have on my to-do list to start a repository to collect some of the work done so far (mpl idioms)
Hi, just in case you find anything useful in it: I wrote https://github.com/pieterdavid/mplbplot to make the plots in my thesis (and LHCb-PAPER-2016-065) with matplotlib from ROOT histograms and graphs (I knew about rootpy, but we already had the pyroot code, so I wanted something more limited in scope to just get the data points and plot them).
hi guys, I made a skeleton at https://github.com/nsmith-/mpl-hep and want to invite people as collaborators to the repository. Let me know if you want in on it. The idea would be to start by compiling a wishlist of what kind of plot styles we want to support, inside the
binder/gallery.ipynb notebook that is also accessible through binder. Feel free to edit and merge the notebooks (although this could get messy..) Then we can take that and make a specification, then start to write an actual python library based on that. Thoughts?
I was just playing with things, and it occurs to me that a lot of our "style" issues for 1D histograms might be solved by using bar plots with the widths parameter. If we want a line style, then we get vertical lines between the bars, but I could imagine this being solved with a new option to the matplotlib method that hides such lines, producing a sky-line effect. Was this already considered somewhere?
The bar chart with no gaps was, but not the new option to Matplotlib. I haven't considered this "just another way to draw a skyline" because the graphics primitives involved are very different. For a bar chart, you make many rectangles—which can be a performance problem if you have a thousand bins—but for a skyline you make a single polyline. It could be that we got used to this style because of the time it takes to draw on a line printer with PAW. (Imagine one of those pen-robot printers trying to draw a thousand rectangles!) Even if that was its origin, I still find the style cleaner and easier to read.
I have no objection to presenting it in the high-level interface as just another drawing option. In fact, that sounds like a good way to get that information from the user. But the implementation is quite different.
Does matplotlib already optimise such things under the hood bonnet ?
If two identical filled paths intersect, is it rendered via the union of such paths, for example?
In other words, Tufte's principle is this: if the plot conveys the same information when you remove a line, then you should remove that line.
Jim posted some examples higher up in this chat (Feb 27 15:49)