# How do you pattern match a DataDistribution

I have a function,

f[dist_, samp_]:=somework[dist, samp]

that I want to return Null or zero if passed a null distribution. I can't find a way to test for a difference between Null and Data distribution: none of ===, =!=, SameQ and == can identify them.

How can I tell if I'm being passed a Null value to handle it correctly?

Seems to be close to what I want.

EDIT: Edited for some clarity.

Here's my function. If probKern is a Null, I want to return a Null, otherwise, I want to calculate the PDF for the sample, with the distribution. I can't quite get a handle on matching the types to make this work though.

calcPDF[probKern_, samp_] :=

calcPDF[probKern_Symbol, samp_] :=
PDF[probKern, samp] /; TrueQ[probKern == DataDistribution]

edit: Okay, I'm still having trouble with this. Here's what I have:

calcPDF[_] := $Failed Given a data distribution and a value it works fine. In[38]:= calcPDF[data, 3] Out[38]= 0.007500611755 Given two numbers, it produces the symbolic value, rather than the -1 I was expect. (The -1 is just a placeholder) In[39]:= calcPDF[3, 3] Out[39]= calcPDF[3, 3] EDIT dec 5th: Running calcPDF by hand seems to result in correct results, but now, when it is threaded over data, the result is a single result of whatever you return in the Except case. There's two sets of distributions I'm running over the sample data, so the final result is a list of two Nulls (or -1's). If you remove the Except case, you get closer to the expected result - two lists 205 elements long, containing PDF's for entries with matching Distributions, and PDF[Null, value] for those without. I'm confused. Clear[calcPDF] (*calcPDF[probKern_, samp_]:=PDF[probKern, samp]; *) calcPDF[probKern_DataDistribution, samp_] := PDF[probKern, samp]; calcPDF[Except[DataDistribution], samp] := Null; (* Pass a list of samples, and a list of distributions's *) applyKernel[lsamples_, kernel_] := Thread[calcPDF[kernel, lsamples]]; (* Pass a list of samples, and a list of two lists of distribution's *) predClass[samples_, kernels_] := Map[applyKernel[samples, #] &, kernels]; predPDF = predClass[predictionSamples[[1]], kernels] - What is a null distribution? – Sjoerd C. de Vries Dec 4 '12 at 21:29 f[dist_, samp_]:=something[dist, samp] /; TrueQ[dist==DataDistribution] Seems to do the trick. if my function isn't passed a valid Data Distribution it falls through to the default case. – Steven Dec 4 '12 at 22:01 Your question is a bit cryptic. Do you mean perhaps that you want to test whether dist is either a distribution that belongs to your null hypothesis or a DataDistribution? For comparing distributions we have DistributionFitTest. – Sjoerd C. de Vries Dec 4 '12 at 22:01 Sorry, I'm looking to test whether dist is a DataDistribution or otherwise (I'm passing null myself). If it's a data distribution, I want to use it in a PDF. Otherwise, I want to pass on a Null value. as it stands when mathematica comes across the Null value, it passes on PDF[Null, some value]. – Steven Dec 4 '12 at 22:31 Steven, you can format blocks of code by selecting them and pressing the { } button above the text box while editor your post. See Editing Help for more. – Mr.Wizard Dec 5 '12 at 6:58 ## 1 Answer I think you are wanting something like the following. For a function that does something with a DataDistribution something else with any other distribution and otherwise fails we would do. f[dist_DataDistribution]:= "do stuff" f[Except[_DataDistribution, dist_?DistributionParameterQ]]:= "do other stuff" f[_]:=$Failed

Now if you want to match a particular DataDistribution, say KernelMixtureDistribution you will need to give it that information in your pattern.

g[_]:= \$Failed

Edit:

I believe your confusion comes from what == is doing in your checks. When you try dist == DataDistribution you are asking if a particular object dist is the symbol DataDistribution. What you really want to know is whether the Head of dist is DataDistribution.

Edit 2:

Lets say you wanted to use this approach to compute CDF for distributions returning -1 if the computation isn't possible. You could set this up as.

cdf[dist_?DistributionParameterQ, x_]:= CDF[dist,x]
cdf[_,x_]:= -1

If you want separate treatment for DataDistribution say you can do that as well. Here I'm telling it to report what type of distribution was used and return the expression and -1 if it isn't a distribution.