Re: Understanding TransformedDistribution
- To: mathgroup at smc.vnet.net
- Subject: [mg131897] Re: Understanding TransformedDistribution
- From: Itai Seggev <itais at wolfram.com>
- Date: Mon, 28 Oct 2013 23:21:15 -0400 (EDT)
- Delivered-to: l-mathgroup@mail-archive0.wolfram.com
- Delivered-to: l-mathgroup@wolfram.com
- Delivered-to: mathgroup-outx@smc.vnet.net
- Delivered-to: mathgroup-newsendx@smc.vnet.net
- References: <20131028044008.90A2A6A13@smc.vnet.net>
On Mon, Oct 28, 2013 at 12:40:08AM -0400, Ernst H.K. Stelzer wrote:
> Through a question on StackExchange I came to know TransformedDistribution, which allows me to sort of develop new distributions based on the ones defined in Mathematica. Instead of using my own functions, which work perfectly well, I now tried to repeat a number of older calculations using e.g. UniformDistribution.
>
> When I combine two via
>
> TransformedDistribution[2 (x - 4/5 y), {x \[Distributed] UniformDistribution[{-1, 1}], y \[Distributed] UniformDistribution[{-(4/5), 4/5}]}]
>
> to generate a kind of annular function and then look at the PDF, I get
>
> PDF[%, x]
>
> \[Piecewise] 1/4 -(18/25)<=x<=18/25
> 1/256 (82-25 x) 18/25<x<82/25
> 1/256 (82+25 x) -(82/25)<x<-(18/25)
> 0 True
>
> However, I expect, what I get when I enter
>
> 2 (PDF[UniformDistribution[{-1, 1}], x] -
> 4/5 PDF[UniformDistribution[{-(4/5), 4/5}], x])
>
> 2 ((\[Piecewise] 1/2 -1<=x<=1
> 0 True
> )-4/5 (\[Piecewise] 5/8 -(4/5)<=x<=4/5
> 0 True
> ))
>
> Does anybody have a clue which mistake I made?
For one thing, your expectation isn't the PDF of anything--it's integral is
2/5, not 1.
PDF is not in any sense a linear operation, because it must always integrate to
1. TransformedDistribution[expr, dists] represnets the distribution of the
values of expr, given that its components are distribution according to dists.
This means your first input represent a distribution ranging from -82/25 (when
x is -1 and y 4/5) to 82/5 (when x is 1 and y 4/5). Since x and y are
independent of piecewise constant distributions, you get a constant expectation
in the "middle" going up from and back down to zero linearly as you go the
maximum negative and positive values.
--
Itai Seggev
Mathematica Algorithms R&D
217-398-0700
- References:
- Understanding TransformedDistribution
- From: "Ernst H.K. Stelzer" <ernst.stelzer@physikalischebiologie.de>
- Understanding TransformedDistribution