Re: [ploticus] New proc boxplot
- Hi Andrew,
sorry for the frustration and for the abruptness of this functionality
change... you might consider sticking with 2.39 for running your existing
proc rangebar code.
one possible way to get it to work w/ 2.40 might be to add an id field
into your input data to make it conform to the new rquirements... even if
it is the same on all rows..
On Tue, 8 Apr 2008, akunn wrote:
> Hello, I am new to this group and am using ploticus for about 2 years
> now. I auto-create ploticus scripts from a database environment to
> produce SVG images.
> I appreciate ploticus in many ways. To me the most important thing is
> that it is script driven and has SVG output. I mainly use bars and -
> the reason why I write - boxplots.
> Steve Grubb wrote something like if there was "any out there using
> boxplots" or similar... I must say, already "functionality was removed
> out of necessity" (Source: Ploticus Handbook) hit me hard.
> Well, here I am - the last and only user of ploticus boxplots. Perhaps
> I am doing something completely wrong .. no entry in this forum
> concerning proc boxplot .. please correct me .. there are some things
> that IMHO just do not work in the new proc .. but nobody seems to
> notice.. looks bad.
> I have 2.40 installed and was able to test the new boxplot today. I
> saw, I will not be able to show outliers and extremes. With this I
> could live, although there are in our special case data driven reasons
> that this feature would give a surplus of information.
> Next I saw that the data needs a break column to fulfill the proc
> processdata requirements. However, without "fields:" attribute, it
> seemes to work. But this rises certain restrictions:
> Without fields: attribute you do not get an id1 ... variable. Placing
> the boxplot on a certain categorical axis position must fail, since
> you only can use the locfield attribute. Without locfield attribute
> everything is nice for the first boxplot, let's say median based. But
> without locfield it is not possible to use proc categories
> slideamount, so painting a second, mean based boxplot, shifted a
> certain amount, fails, too.
> If you now try in this setting to set the mediansym to anything as
> specified, you will find the dot in the left up corner of the area
> rectangle. If the attribute is not mentioned, a line is drawn
> correctly. This without raising an error or warning.
> If you try to set the meansym, that normally would "draw a mean point
> within the boxplot" (like this getting back some of the rangebar
> features) it gives an error message like "xxx needs 7 variables to
> show this, including mean."
> I have tried around for a whole day now and I did not find any
> solution to draw two boxplots without having a break column. The
> former barloc gave the possibility to absolutely position a horizontal
> boxplot at a certain category position on the y axis. The new locfield
> refers to a column that - in my case - does not exist.
> Now that you know, that there is one out here using boxplots - I would
> really be glad to get, let's say, a development path for proc boxplot.
> And, btw, are there other procs, that have in your opinion "functions
> with no necessity" and that could be subject to a drastical change? ;-)
> Yours, Andrew
Stephen C. Grubb scg@... x-6633
Scientific Software Engineer, The Jackson Laboratory
600 Main Street Bar Harbor, Maine 04609 USA