38 ms·
I commend the effort to teach statistics to python programmers, but I'd recommend interfacing with R for creating the figures. R will create beautiful figures b
by pama 16y ago
I commend the effort to teach statistics to python programmers, but I'd recommend interfacing with R for creating the figures. R will create beautiful figures by default; it can also handle sophisticated stats, should the need arise. The current figures are filled with unneeded inwards-facing ticks that often overlap with data, and have tick and axis labels of inconsistent font sizes. The negative numbers in the y-axis in Fig 2.3 and x-axis in Fig 4.6 and 7.1 don't display properly, and the presentation of the data in Fig 3.1 is too noisy to deliver a useful message.
- GregBuchholz 16y agoI find that ticks aren't enough to aid in real problems, and I always turn on the the grid with 'grid(True)' in matplotlib. Also, the negative number problems must be some kind of anomaly; I've never seen that problem, and I've used several revisions of matplotlib every day for the past couple of years.
- pama 16y agoOut of curiosity, why do you need a grid? (I need very few ticks myself and almost never a grid. I've found that if the visual message isn't clear enough or needs a different quantitative analysis or display, I'll request the raw data.)
- GregBuchholz 16y agoI'm almost always interested in the actual values of things when debugging a part our systems at work. I need to know that the generator exceeded 3kW, or the PDF of the voltage was above 4V, or things went hay-wire at 11:31AM. It is easier to tell at a glance what's happening on a plot if there is a grid.
- araneae 16y agoI think the point is to learn statistics using an easier language than R, ugly figures be damned.
- rch 16y agoWell, the R syntax is more intuitive than statistics itself. Then again, what isn't? Really though- rpy2 is a nice enough interface to R, and things like MCMCPack make R worth the effort.
- phren0logy 16y agoI see this as a real issue. R is useful but sorta crappy as languages go. I have hopes for Incanter [http://incanter.org/ http://incanter.org/] (a clojure-based stats environment), but it looks stalled out vs dead in the water. Even still, for basic stats it's pretty viable. I would love to see a stats environment based on a sane, full-featured language that was ready to take advantage of multiple cores and GPU computing. Clojure has all the building blocks, but it's not built yet.
- araneae 16y agoI have to confess that whenever I do any statistics (which isn't much, these days) I struggle with R for a bit and then say "screw it" and use MatLab.
- pama 16y agoQuick R, helped several of my friends try R: http://www.statmethods.net/ http://www.statmethods.net/ Once you're hooked, go to the R inferno: http://www.burns-stat.com/pages/Tutor/R_inferno.pdf http://www.burns-stat.com/pages/Tutor/R_inferno.pdf
- JoshCole 16y agoI had been planning to work on Incanter over the summer as part of Google Summer of Code, but things didn't pan out. The organization promoting the project wasn't accepted as a mentoring organization. Now I have to find another internship.
- graywh 16y agoWhat about R do you think makes it hard?
- Jach 16y agoThere's only a million other plotting libraries you can use with Python if you don't like what pyplot spits out. I think R's pie charts are hideous myself. (And I wouldn't ever use R's default output for a BI chart.)
- ylem 16y agomatplotlib has an excellent plotting package and I've been using it in publications.
- ZeroGravitas 16y ago"One of the reasons we are using a general-purpose language rather than a stats language like R is that for many projects the "hard" part is preparing the data, not doing the analysis." quote from week 2 of the accompanying lecture notes: https://sites.google.com/site/thinkstats2010b/lecture-notes/lecture-2 https://sites.google.com/site/thinkstats2010b/lecture-notes/...
- woodson 16y agoI find preparing the data is actually easier in R than in something like python, but that certainly depends on type and amount of data. And, of course, each to their own..