From: Benjamin Root <ben.root@ou...>  20120712 18:52:22

On Thu, Jul 12, 2012 at 1:58 PM, Daπid <davidmenhur@...> wrote: > I don't know if there is any reason for not having it, but as a > workaround, you could use np.hist to get the data (syntax is the same > as mpl.hist and returns the same numbers, but without drawing) and > then renormalise and plot with mpl.bars. > > On Thu, Jul 12, 2012 at 4:42 PM, Alan G Isaac <alan.isaac@...> > wrote: > > This is essentially a query about why certain histogram types > > are not offered. I can see two possible answers: haven't gotten > > to them, or, don't want to offer them (e.g., they're bad practice). > > > > I will choose Stata as a point of comparison. > > http://www.stata.com/help.cgi?hist > > The types are density, fraction, frequency, and percent. > > Frequency corresponds of mpl's normed=False. > > Density corresponds of mpl's normed=True. > > Today I wanted the 'fraction' type, but mpl did not offer it. > > (Note that because of other elements of the graph, hacks like > > replacing the ticklabels won't work nicely.) > > > > If there is not sentiment against offering these types, > > I suggest that the `normed` keyword accept strings, > > including "fraction" and "percent", and that `hist` > > be extended to produce these types. > > > > Cheers, > > Alan Isaac > > > Alan, It is tricky to balance functionality, conciseness, and usability. It is both a shortcoming and a strength that we strive to provide a nofrills library that gives the user plenty of rope, while still trying to anticipate most users' needs so that they can get stuff prototyped and working. For most other plots, one would simply convert the data units themselves prior to plotting. However, hist() is a different beast from say, plot() or scatter(), because hist() inherently performs calculations upon the input data and displays the results of those calculations. Therefore, one can't simply perform a conversion on the data prior to calling hist() like one would for other functions. In that light, your idea makes a lot of sense and could essentially be considered as a way to "pretty print" what the graph would normally look like if normed=True. In other words, having normed='percent' would essentially perform the calculations exactly as it would if normed=True, but would multiply the ycoordinate system by 100. Is that what you are looking for? Cheers! Ben Root 