Comparing Numbers - improving on log relative error
- To: mathgroup at smc.vnet.net
- Subject: [mg27279] Comparing Numbers - improving on log relative error
- From: "mvyver" <mvdv at bigfoot.com>
- Date: Sun, 18 Feb 2001 02:52:13 -0500 (EST)
- Organization: The University of Western Australia
- Sender: owner-wri-mathgroup at wolfram.com
Hi,
This may be trivial... the idea is to take two numbers, or lists of numbers,
sumbit them to a function that returns a set of integers indicating the
numer of digits in agreement. An eaxample should be clearer:
c = {-0.01683787816107418, 0.22080842442483295, 1310.3262295713917};
est = {{-0.01821756936118390, 0.2208605273841613, 1310.3353890004764},
{0.017177028591333965, 0.22112030617117995, 1310.3270577680291},
{-0.01715869524044578, 0.22093374833870505, 1310.3266833564685},
{-0.016406423628831955, 0.2209052735237087, 1310.3260455795904}};
Comparing c to each of the lists in est should return something like - this
is by hand so there may be an error :)
result={{1, 3, 5}, {0, 2, 6}, {1, 3, 7}, {1, 3, 7}}
one approach is the log relative error:
LRE = -Log[10, Abs[(# - c)]/Abs[c]] & /@ est
which is off in some cases:
lreresult={{1.08651, 3.62715, 5.15551}, {-0.305382, 2.85003, 6.19925},
{1.72003, 3.24598, 6.46053}, {1.59135, 3.35792, 6.85258}}
I tried applying RealDigits to each of the numbers, but then hit a snag in
counting the matching elements accurately. If it make things easier the
sign checking I can do in a seperate step.
A more elaborate approach would indicate the bias in est relative to c, but
expect this may be quite tricky, eg. the result would then be:
result={{-1, 3, 5}, {-1, 2, 6}, {-1, 3, 7}, {1, 3, -7}}
Any suggestions/help appreciated.
Mark Van De Vyver
Dept Acc & Finance
University of Western Australia
Perth