PuffinPlot issue tracker

Bug: a88/8c9

ID : 8c98723e-259b-45eb-b7a9-6a04147a32a8
Short name : a88/8c9
Status : open
Severity : minor
Assigned :
Reporter : Pontus Lurcock <pont@talvi.net>
Creator : Pontus Lurcock <pont@talvi.net>
Created : Tue, 28 Jul 2009 02:31:03 +0000
Target : 1.5
Summary : Automatic PCA selection [94]

Comment: --------- Comment ---------
ID: 505ae411-622f-4f05-95b5-78f3051c9bd5
Short name: a88/8c9/505
From: Pontus Lurcock <pont@talvi.net>
Date: Tue, 28 Jul 2009 02:31:03 +0000

Christian claims that Gary Acton has a script which does this, and
has asked for it. An algorithm for this is also described in the
Kirschvink paper.

Comment: --------- Comment ---------
ID: 7d561e12-f66f-4c25-9075-4094ecf58fae
Short name: a88/8c9/7d5
From: Pontus Lurcock <pont@talvi.net>
Date: Thu, 09 Feb 2012 22:41:11 +0000

Gary Acton has emailed me the code for this (in Fortran),
I think.

Comment: --------- Comment ---------
ID: 5823484d-7ea6-48ed-adcd-b53020b2d220
Short name: a88/8c9/582
From: Pontus Lurcock <pont@talvi.net>
Date: Thu, 10 May 2012 14:51:59 +0000

Also mentioned by GA in review comments on the paper:
‘For batch processing (like the many thousands of intervals measured along
U-channels), an automated PCA that recognizes and removes outliers would be
useful. That is the primary feature that ZPLOTIT has that PuffinPlot does not
and is the main reason I wrote ZPLOTIT. The outlier removing capability would
be easy to incorporate. In ZPLOTIT the user defines a minimum number of steps
to use in the PCA and a range of demagnetization steps over which to search.
If the range contains more than the minimum required steps, the program seeks
outlier steps, which are steps that degrade the PCA line fit significantly.
The PCA is then computed with and without outliers and the anchored and
unanchored PCA is provided for comparison. In PuffinPlot, one could
accomplish this but currently would have to do so for each individual sample,
which is not feasible when dealing with hundreds of meters of U-channels.’
This is presumably the algorithm implemented in the Fortran code,
and also looks similar (or possibly even identical) to the Kirschvink

Comment: --------- Comment ---------
ID: baac1ead-6334-4e7e-8092-d2f539bc5bae
Short name: a88/8c9/baa
From: Pontus Lurcock <pont@talvi.net>
Date: Thu, 10 May 2012 14:52:49 +0000

Provisionally assigning to the ‘revised paper’ release, though
I may have to drop it if time runs out.

Comment: --------- Comment ---------
ID: 283d46a1-16a1-4594-83a0-7db74a7c3908
Short name: a88/8c9/283
From: Pontus Lurcock <pont@talvi.net>
Date: Tue, 22 May 2012 21:21:30 +0000

There won't be time to implement this for v1.01.