Re: Replace Table[Sum...]...] with Functional code
- To: mathgroup at smc.vnet.net
- Subject: [mg98751] Re: Replace Table[Sum...]...] with Functional code
- From: green tea <cetech07 at gmail.com>
- Date: Sat, 18 Apr 2009 03:36:20 -0400 (EDT)
- References: <grn1l3$n0t$1@smc.vnet.net>
On Apr 10, 5:57 pm, Andreas <aa... at ix.netcom.com> wrote:
> I use the following two functions to analyze time series:
>
> mm[data_, period_] := Table[Sum[(Log[data[[t]]] - Log[data[[t - i]]]) / Sqrt[i], {i, 1, period}] / period, {t, period + 1, Length[data]}]
>
> mr[data_, period_] := Table[Sum[Log[data[[t - i]] / data[[-i + t - 1]]] * (Sqrt[i] - Sqrt[i - 1]), {i, 1, period}],{t, period + 2, Length[data]}]
>
> They have a similar structure, Table[Sum[...]...], which sums a bunch of stuff from the time series at each point in the series.
>
> I'd like to convert them into functional programing constructs, but as the increments in the Table function come into play in the Sum, I haven't been able to figure out a way to do it.
>
> I've tried Map with Partition and a bunch of other things, but I could use some pointers.
>
> Any help much appreciated.
ListCorrelate is very fast, but
generalized ListCorrelate like "ListCorrelate[{1,1},data,{1,-1},
1,Times,Log[#2/#1]&]" is terribly slow.
I don't know why.
instead use these codes.
Log[data/RotateRight[data]]
Log[Rest[data]/Most[data]]
Log[Ratios[data]]
etc.
ListConvolve's computational compexity is O(n log(n)),
and it does not create intermediate list structure like Partition and
Dot combination.
so it is more suitable for this time series analysis.
mm[data_, period_] :=
With[{kernel = (period Sqrt[Range[1., period]])^-1, list = Log
[data]},
list[[period + 1 ;;]] Total[kernel] - ListConvolve[kernel, list[[;;
-2]]]]
mr[data_, period_] :=
With[{kernel = Differences[Sqrt[Range[0., period]]], list = Log[Ratios
[data]]},
ListConvolve[kernel, list[[;; -2]]]]