tag:blogger.com,1999:blog-9959776.post2961899183159798155..comments2024-02-15T04:42:41.606+00:00Comments on James' Empty Blog: UghJames Annanhttp://www.blogger.com/profile/04318741813895533700noreply@blogger.comBlogger13125tag:blogger.com,1999:blog-9959776.post-7490243714583573962007-08-23T14:37:00.000+01:002007-08-23T14:37:00.000+01:00I doubt you missed anything because I didn't see a...I doubt you missed anything because I didn't see anything either.<BR/><BR/>Are they using it? Who knows? I happen to be interested in CPDN sampling strategy and from time to time have a go at trying to get more information made available to participants<BR/><BR/>http://www.climateprediction.net/board/viewtopic.php?p=67053&highlight=#67053<BR/><BR/>I seem to end up with general replies about what would be a 'good feature' but isn't actually in use. A reasonable conclusion might be that the strategy is still fairly dumb/random/non-optimised and maybe they will think about clever techniques sometime.crandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-22181217688816662932007-08-23T09:47:00.000+01:002007-08-23T09:47:00.000+01:00I don't think Piani et al directly proposed any me...I don't think Piani et al directly proposed any method for sampling parameters. Of course it's not out of the question that I missed something...are cpdn actually using it to design their sampling strategy?James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-18172755238693631012007-08-22T13:45:00.000+01:002007-08-22T13:45:00.000+01:00Thank you for the reply.My reaction is that you us...Thank you for the reply.<BR/><BR/>My reaction is that you used 'large ensemble' and 'generate a smaller ensemble' and I would point out that large and small are relative terms. If CPDN is ongoing (and more CPDN slab models have recently been released) then perhaps you have a large ensemble that is going to get larger. The issue of how to expand the sample size could still involve working out how to 'generate a smaller ensemble' that provides better information couldn't it? <BR/><BR/>So I struggle to see why you conclude it is not relevant. Different scale to what you were thinking maybe.<BR/><BR/>crandlescrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-66888486798787065012007-08-22T12:37:00.000+01:002007-08-22T12:37:00.000+01:00Oh, now I see what you re getting at. Not an unrea...Oh, now I see what you re getting at. Not an unreasonable question, but I don't think their work is really relevant to what we were talking about - that's not intended as a snipe, just that they are analysing a large ensemble that already exists, rather than trying to efficiently generate a smaller ensemble thst samples the posterior (of an application of Bayes' Theorem), which is what I've been trying to do.James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-56657441269765482392007-08-22T10:00:00.000+01:002007-08-22T10:00:00.000+01:00Well yes but your priors paper has been discussed ...Well yes but your priors paper has been discussed enough here and maybe not everything revolves around your priors paper.<BR/><BR/>You have a new paper which seemed like a round up of various sampling strategies and their efficiency for gaining probability information. This didn't seem to mention Piani et al (2005)'s use of transfer functions. Is it reasonable for me to be asking why there was no mention of it and whether this technique has implications for sampling strategy?<BR/><BR/>crandlescrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-64981661151442577172007-08-22T04:44:00.000+01:002007-08-22T04:44:00.000+01:00"your paper is the last word on the subject"Certai..."your paper is the last word on the subject"<BR/><BR/>Certainly not, although the fact that (AFAIK) no-one has attempted to improve, update or challenge what we have said in the 18 months since the GRL paper was published may give pause for thought...<BR/><BR/>Of course estimates can be expected to improve as we learn more about the climate system. What I do claim about our papers is that they present fundamental principles that any future estimate must take into account if it is to be considered credible (or even meaningful).James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-26887330997995729122007-08-21T20:37:00.000+01:002007-08-21T20:37:00.000+01:00Sorry. 'their' probably was a bit vague.I think I ...Sorry. 'their' probably was a bit vague.<BR/><BR/>I think I was just talking about Piani et al and CPDN's use of the technique in that paper:<BR/><BR/>http://www.climateprediction.net/science/pubs/Piani_GRL.pdf<BR/><BR/>though to be honest in writing the comment, I didn't reread the paper (which I don't think I fully understand). I was just reviewing <BR/><BR/>http://www.climateprediction.net/science/pubs/cpdn_for_oucs.pdf<BR/>(pages 41 to 44)<BR/><BR/>Anyway your reply seems to indicate you think it is a backward step and your paper is the last word on the subject (even though you have indicated otherwise elsewhere).<BR/><BR/>Might an alternate view might be that we need to properly work out the probability implications from one source before we try to combine pdfs using different sources?<BR/><BR/>crandlescrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-78861964542678882242007-08-21T12:12:00.000+01:002007-08-21T12:12:00.000+01:00Um, that's a bit cryptic. "their use of transfer f...Um, that's a bit cryptic. "their use of transfer functions" refers to another paper in the issue? My hard copy of the issue is doing the rounds at the lab and I've not actually read it properly.<BR/><BR/>But regardless, the whole scene seems pretty farcical in the way people keep footling around with cod philosophy rather than actually dealing with the arguments raised in our GRL (multiple constraints) and arxived (priors) papers.James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-52981891427477706402007-08-20T11:20:00.000+01:002007-08-20T11:20:00.000+01:00Thank you for the reply.Does their use of transfer...Thank you for the reply.<BR/><BR/>Does their use of transfer functions al la Piani et al tell you anything about their sampling strategy or rather how you think it should develop?<BR/><BR/>Is the Piani et al method of arriving at the same pdf for sensitivity whether you estimate it directly or via lamda reasonable or does this still have a significant problem from using a linear prior?<BR/><BR/>Regards Chriscrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-25599561710094720162007-08-20T00:07:00.000+01:002007-08-20T00:07:00.000+01:00To be honest it was thrown together in a hurry and...To be honest it was thrown together in a hurry and designed to be rather bland and uncontroversial - at the time I had had enough of rejections!<BR/><BR/>It then turned out that none of the papers in that issue got properly reviewed (as you may guess from the contents of some...) so I could have written pretty much anything...ho hum.<BR/><BR/>Anyway, it looks like you got the gist of it clearly enough.James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-70772019120448521512007-08-18T17:46:00.000+01:002007-08-18T17:46:00.000+01:00Your paperhttp://www.jamstec.go.jp/frsgc/research/...Your paper<BR/>http://www.jamstec.go.jp/frsgc/research/d5/jdannan/philtrans_final.pdf<BR/><BR/>sounds interesting. Whether I am up to understanding it, I am not very sure.<BR/><BR/>See http://www.climateprediction.net/board/viewtopic.php?t=7453<BR/><BR/>crandlescrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-57644526303664658052007-08-17T02:03:00.000+01:002007-08-17T02:03:00.000+01:00Chris,I don't think it's worth going as far as try...Chris,<BR/><BR/>I don't think it's worth going as far as trying to include a Bayesian primer in each set of documentation, but I did write to the responsible parties and it looks like both sets of documentation will get corrected, so at least they won't actively mislead people in future.<BR/><BR/>As for "worse use of your time", well that depends on whether I'm writing for my personal entertainment, to entertain my reader, as a handy notebook for future reference, or to change the world :-)James Annanhttps://www.blogger.com/profile/04318741813895533700noreply@blogger.comtag:blogger.com,1999:blog-9959776.post-39850742335476375022007-08-15T21:47:00.000+01:002007-08-15T21:47:00.000+01:00HiSo are you going to be writing to the producers ...Hi<BR/><BR/>So are you going to be writing to the producers of this errant documentation? Will you suggest improvements to the documentation that specifically mention that:<BR/><BR/>The probability of data given a hypothesis (P(D|H)) should not be confused with probability of hypothesis given the data P(H|D). Also that Bayesian analysis is almost certainly required for P(H|D) together with its need for a prior distribution.<BR/><BR/>It may take a long time for that sort of documentation to be noticed and help your cause, but it surely cannot hinder. (I cannot imagine that it would be a worse use of your time than repeating the same thing here all the time.)<BR/><BR/>crandlescrandleshttps://www.blogger.com/profile/15181530527401007161noreply@blogger.com