Anal1a, in terms of clumping, doesn’t look very good, at least not anymore. That was not previously the case, but I had revised my algorithm to make it symmetrical as per the insructions of a computing statistician here at the University of Sussex. He claimed that the Procrustes Rotation needed symmetric data and my nonsymmetric data, where Doc1 vs Doc2 didn’t have the same similarity as Doc2 vs Doc1, was not going to work. That change has, I believe, altered the efficacy of the algorithm and things are no longer clumped together as promisingly as they were previously. The clumps should be a two- or three-letter short code followed by a digit. Therefore, ac1 and ac2 belong together. Pl1, pl2, and pl3 belong together, and so on. The clumping is significantly better in the already symmetric cosine normalization algorithm (anal2a). The two speech processing documents are clumped together (sp1 and sp2), all of the Power PC and G4 documents are together (pp1, pp2, g4c), and the three Pine Lake tornado stories are clumped far away from everything else (which is all computer-related) and together on their own. Excellent clumping, in fact. So the hypernym hypothesis looks like, on these short documents, it is working well with cosine normalization.

Here’s the final bit of loveliness: comparing one MDS cluster diagram against another. MDS output is mapped to the vector space independently. That is, the same data will produce the same visualization or mapping, but different data is mapped to a different vector space, so you cannot just compare one MDS matrix to another directly. That is where Procrustes Rotation comes in. It applies a series of intelligent matrix transformations, trying to map the second vector matrix onto the source vector matrix. As a side benefit, essential in my case, it always provides a fitness measure to tell you how close the two were. on a scale of 0 to 1. So these two, as you can see (see above image), even after the transformations, were not that close together. As it happens, though, this is not particularly useful information to know. I am currently more interested in assessing how close the two algorithms are to human classifiers.

This recent success gave us plenty to discuss, particularly with respect to metric and non-metric data. The MDS community calls source data metric when the similarity or dissimilarity data is symmetric. That is, the value at row 2, column 1 is the same as the value at row 1, column 2. Classical multi-dimensional scaling (MDS) is designed to only work with metric data. SPSS includes the ALSCAL and PROXSCAL MDS algorithms which can work with non-metric data, but MATLAB’s classical MDS does not because it treats things as Eucledean distances–another reason why I had to alter the Anal1a algorithm. The primary reason I now had metric data for everything, however, was because the computing statistician had told me I needed it for the Procrustes. Hawever, as we were examining my output, it occurred to me that Procrustes did not really care if the data was symmetric, so long as the dimensions of the data were the same (the same number of rows and columns). Which leads us to question whether the application of the method is statistically sensible or not. To that end, I need to track down a new computing statistician and perhaps a mathematician and discuss the process with them. My original computing statistician has retired.

Earlier I said that comparing one machine to another, to see how they fit is not useful information, but what would be interesting is to prepare a matrix of all the possible combinations of human judgements, cosine normalization, and weird formula:

cosine   wrd form.   human
cosine (anal2a)		x
weird formula (anal1a)           x
human                                        x

So that is my task for my next meeting (on the 16th of June). Before then, I need to figure out how to get MATLAB to take multiple tables as data. In SPSS, I could paste in several tables (representing all of the people’s individual data, for example) and it would work with that. That is necessary in order to aggregate the peopel to do the comparison. Onward ho, then! Progress at last!

Share/Save

Doc1	Doc2	Doc3	Doc4
Doc1	100	76	18	91
Doc2	76	100	22	35
Doc3	18	22	100	65
Doc4	91	34	65	100

This is not too dissimilar from the labelled diagram, part of the MATLAB documentation on data importing. Except that, if you look at the table below it, which describes which functions to use, they don’t have a function with a similar example to their labelled diagram. Early on I thought I should be able to use dlmread, which allows you specify rows/columns for starting points or a range. My idea was just to have a range which excluded the non-numeric troublesome labels. No matter what I did, though, I could not get it to work. It was frustrating, because I could paste the data into the Import Wizard and that could handle the data fine. I wrote people, I researched on the web, and I tried all sorts of things.

Eventually, I came full-circle back to dlmread and experimented by making a small data file with unrelated data in it. That worked fine. So I then copied half of one of my data tables into the test file and tried that. That also worked fine. I copied the whole data table into the test file and used dlmread on it. It worked fine! What was the difference between the two identical data files other than their filenames? When I uncovered the answer to that, I kicked myself. My data files were generated years ago and stored on my Mac OS 9-based laptop. My laptop and the data have since migrated to Apple’s swoopy BSD-based UNIX goodness and that’s the environment that MATLAB runs under. So… Have you guessed the problem? Yes, it was linefeeds! The data files had original Mac linefeeds and MATLAB wanted UNIX linefeeds. D’oh! It just goes to reaffirm that the things you don’t see can really hurt you.

Once that was solved, work proceded rapidly apace as I was now able to finish automating the whole comparison process from start to finish.

function  [Anal1Raw, Anal2Raw, Anal1MDS, Anal2MDS, fit] =
processEinCiteData(firstFile, secondFile, runName, labels)
% Read in the similarity matrices from the two data files
Anal1Raw = dlmread(firstFile, '\t', 1, 1);
Anal2Raw = dlmread(secondFile, '\t', 1, 1);
% Set up default document name labels if we didn't get any
if nargin < 4
labels = {'g4c', 'pp1', 'pp2', 'msc', 'pl1', 'pl2', 'pl3', 'sp1', 'sp2', 'ac1', 'ac2', 'bws'};
if nargin < 3
runName = '';
end
end
% Set up labels for the filenames
fileName1 = regexprep(firstFile, '\..*$', '');
fileName2 = regexprep(secondFile, '\..*$', '');
% Convert the similarity data to numbers below 1 for use in MDS
Anal1Raw = abs(100 - Anal1Raw)
Anal2Raw = abs(100 - Anal2Raw)
% Calculate the MDS and prepare a diagram showing the
% clusterings for the first document
[Anal1MDS, eigvals] = cmdscale(Anal1Raw);
figure(1);
plot(1:length(eigvals),eigvals,'bo-');
graph2d.constantline(0,'LineStyle',':','Color',[.7 .7 .7]);
axis([1,length(eigvals),min(eigvals),max(eigvals)*1.1]);
xlabel('Eigenvalue number');
ylabel('Eigenvalue');
plot(Anal1MDS(:,1),Anal1MDS(:,2),'bo', 'MarkerFaceColor', 'b', 'MarkerSize', 10);
axis(max(max(abs(Anal1MDS))) * [-1.1,1.1,-1.1,1.1]); axis('square');
text(Anal1MDS(:,1)+1.5,Anal1MDS(:,2),labels,'HorizontalAlignment','left');
hx = graph2d.constantline(0,'LineStyle','-','Color',[.7 .7 .7]);
hx = changedependvar(hx,'x');
hy = graph2d.constantline(0,'LineStyle','-','Color',[.7 .7 .7]);
hy = changedependvar(hy,'y');
title(['\fontname{lucida}\fontsize{18}' fileName1 ' MDS']);
xlabel(['\fontname{lucida}\fontsize{14}' runName ' on ' date], 'FontWeight', 'bold');
% Calculate the MDS and prepare a diagram showing the
% clusterings for the second document
[Anal2MDS, eigvals] = cmdscale(Anal2Raw);
figure(2);
plot(1:length(eigvals),eigvals,'rd-');
graph2d.constantline(0,'LineStyle',':','Color',[.7 .7 .7]);
axis([1,length(eigvals),min(eigvals),max(eigvals)*1.1]);
xlabel('Eigenvalue number');
ylabel('Eigenvalue');
plot(Anal2MDS(:,1),Anal2MDS(:,2),'rd', 'MarkerFaceColor', 'r', 'MarkerSize', 10);
axis(max(max(abs(Anal2MDS))) * [-1.1,1.1,-1.1,1.1]); axis('square');
text(Anal2MDS(:,1)+1.5,Anal2MDS(:,2),labels,'HorizontalAlignment','left');
hx = graph2d.constantline(0,'LineStyle','-','Color',[.7 .7 .7]);
hx = changedependvar(hx,'x');
hy = graph2d.constantline(0,'LineStyle','-','Color',[.7 .7 .7]);
hy = changedependvar(hy,'y');
title(['\fontname{lucida}\fontsize{18}' fileName2 ' MDS']);
xlabel(['\fontname{lucida}\fontsize{14}' runName ' on ' date], 'FontWeight', 'bold');
% Apply Procrustes to the two MDS results to map them
% into the same vector space and prepare a plot of the
% result
[fit, Z, transform] = procrustes(Anal1MDS, Anal2MDS);
figure(3);
plot(Anal1MDS(:,1), Anal1MDS(:,2), 'bo','MarkerFaceColor', 'b', 'MarkerSize', 10);
hold on
plot(Z(:,1), Z(:,2), 'rd', 'MarkerFaceColor', 'r', 'MarkerSize', 10);
hold off
text(Anal1MDS(:,1)+1.5,Anal1MDS(:,2), labels, 'Color', 'b');
text(Z(:,1)+1.5,Z(:,2),labels, 'Color', 'r');
xlabel(['\fontname{lucida}\fontsize{14}' runName ' on ' date], 'FontWeight', 'bold');
ylabel(['\fontname{lucida}\fontsize{14}' 'fit = ' num2str(fit, '%2.4f')], 'FontWeight', 'bold');
titleStr = ['\fontname{lucida}\fontsize{18}' fileName1 ...
' compared to ' fileName2];
title(titleStr, 'HorizontalAlignment', 'center', ...
'VerticalAlignment', 'bottom');
legend({firstFile, secondFile}, 4);

At the end, I had a quantitative number, the degree of fit, between two diagrams after applying the Procrustes Rotation to them. Finally! On a whim, I fed in the same data table as both arguments to my comparison program. That is, I compared the same data file to itself. My hypothesis was that the resultant degree of fit should be either 0 or 1 (depending on which the fitness was measured). Much to my surprise, no matter which data file I used, the result was never 0 or 1. My previous Procrustes Analysis code was taken from some sample code in the MATLAB documentation and looked like: [D,Z] = procrustes(Anal1aMDS, Anal2aMDS(:,1:2)); That last bit in () is some kind of MATLAB scaling, which, being a novice to MATLAB, I didn’t realize. So, in fact, my two diagrams weren’t the same which is why I wasn’t getting a 100% degree of fit. I do not want to say how long it took me to narrow that down. Once I did, though, it looked like I was basically set and I was able to quickly produce some comparisons between my “weird” half-baked metric and the cosine normalization one. One small step for EinKind.

This is a delayed entry from May 12th, 2004.

Share/Save

Ciphers blowfish-cbc, aes128-cbc, 3des-cbc, blowfish-cbc, cast128-cbc, arcfour, aes192-cbc, aes256-cbc
Compression yes
ForwardX11 yes
Protocol 2,1
RhostsAuthentication no
RSAAuthentication no
UseRsh no

Note: It may be necessary to take out the extra spaces between items in the Ciphers list.

That, of course, assumed I had already had an entry in my ~/.cshrc file to the effect of:

setenv DISPLAY :0.0

Then, I can simply use a variant of the following to, for example, launch dtterm:

ssh -X eingang@machine_name.com /usr/dt/bin/dtterm

MATLAB, unfortunately, doesn’t seem to work at all. It used to generate all kinds of font errors looking for various Sun fonts, but it did launch. Now it launches with all the font errors (see sample below), shows the graphical about box and then we’re in the command line environment instead of the GUI development environment.

Font specified in font.properties not found [ urw itc zapfdingbats medium r normal * %d * * p * sun fontspecific]
Font specified in font.properties not found [ urw itc zapfdingbats medium r normal * %d * * p * sun fontspecific]

Note: All the ‘-’ characters appearing inbetween [] above have been changed to spaces to render better in HTML.

I’ve mailed the Mac technical support person to see if they have any ideas as to why it no longer works and how it might be possible to fix the font errors.

Share/Save

As I haven’t met with Thufir since receiving the e-mail, I haven’t discussed it with him to see if he’s willing to cough up £525 (~930 US/780 €) plus £210 (~370 US/310 €) for the software for me. The lab manager stopped by my office today to see if somebody was going to pay for it and we entered into a discussion about the licensing arrangements. The academic license is only available for installation on university-owned equipment. I don’t have any university-owned equipment. He doesn’t want to be breaking the law by violating the terms of the licensing agreement which is fair enough. That does leave me in somewhat of a quandry: I don’t qualify for the student edition; I’d be in violation of the license for an academic version; and I can’t afford the £1625 (~2900 US/2400 €) commercial base software price plus an additional £600 (~1050 US/895 €) for the toolkit I need.
Why do I need my own copy? Can I live without my own copy? I can possibly do without my own copy as I can run it on campus, but I’m not on campus all the time. I do return to Canada for extended periods of time and we are investigating an option of moving back to Canada and having me commute out every quarter or so for a month as that would be cheaper than actually living here. Just last year I spent 6 months in Canada. Even if I am here, it’s a problem if I just want to work from home. It would therefore be nice to have my own copy, but at what price?
I’ve told the lab manager that I’ll get back to him. I’m meeting with Thufir on Friday. I also e-mailed MathWorks to ask about my situation and licensing arrangements for it. It does seem strange that their academic license is restricted by machine ownership rather than by reported use for the software. Their customer service auto-responder promises a response within 24 hours, but so far all I’ve received is the automatic response and a response from their database that my e-mail address has been updated in my profile. I have no idea what that’s about. Maybe tomorrow will be better.

Share/Save

David L. Jones had a series of MATLAB pointers which included links to toolboxes for non-metric multidimensional scaling. The latter toolkit, developed by Mark Steyvers, doesn’t come with any documentation and includes some DLLs, so I wonder if only works in Windows somehow? I couldn’t find any other reference to it on the web.
I was waiting for the Mac support person to come install a new version of MATLAB for me. The demo installation and toolkits I installed last fall have long since expired. I’m also still waiting to hear back from the UNIX software support people in the department about acquiring one of the pool licenses for use with a copy of MatLab on my Macintosh off campus. Latish on in the day, I found the Mac support person and acquired a valid license file. It didn’t work right off the bat. I had to edit the file and change the linefeeds from Macintosh ones to UNIX ones. After that, it worked great and it looks fantastic. So I should be able to start doing something with that soon. It also works from home, surprisingly enough, as long as I have an Internet connection, so that will be quite convenient. Hurrah! I am moving ahead.

Share/Save

Despite my pitiful progress on the items on my list, I hadn’t been completely idle. I was able to report that I had been investing time in helping coordinate and set-up the new FirstClass intranet to be used by the IDEAs Lab. I have also been providing some advice to Julie Coultas on creating a web-site specification for her EduServ project. These activities do not directly further my Ph.D. work, but Thufir pointed out that they serve to help relieve the intrinsic isolation of my position in the group, as I am not working on a project with others and my work is not very directly related to most of the Lab’s projects. I must be careful, however, not to spend the majority of my time on Lab-related work.
On my list of things still to do is to sort out my MATLAB problems. Prior to leaving for Canada last October, I was in correspondence with Dr. Tom Browne, who is responsible for academic license management here at the University of Sussex. While the software can be run directly on campus from various UNIX boxes, copying and pasting data/diagrams between the UNIX machine and the Mac isn’t particular convenient. Sussex maintains a pool of MATLAB licenses. Purchasing individual licenses and toolkits is very expensive, so I have been trying to acquire the use of a license for use on my Macintosh. I need to follow up and see what the status of this last was.
One of the things Thufir is concerned about is that I have been basically stuck at the same point of my Ph.D. for several years now. Of course, that has not been helped by my lengthy absences due to work and my poor health over the past few years. It also has not been helped by the fact that the analysis I need to do is quite tricky and not understood by many people. Thufir admitted that the statistics was beyond him. Imagine a supervisor not knowing everything, but he was happy to admit he’s not the end-all, be-all for everything. He tried forwarding me to various people and I spent ages asking people for advice and help all over the Internet and locally before I chanced across something on my own. The other factor that has been impeding the analysis is that statistics is not my strong point. I do not enjoy doing this analysis, yet it is hanging over my head like the sword of Damocles. He worries the analysis tarpit is making my Ph.D. a chore and leading to anxiety and a constant circling without any substantial progress. He is looking for ways to help me break the “logjam.” He suggested two things. One thing was to look for something “fun” that I could start work on now, which might help rekindle my enthusiasm for my project. Another related idea is to see if there is something I can do which does not hinge upon the results of the analysis, perhaps, for example, some of the writing. I promised to look at that as I continue my work on updating my timeline.
Finally, we discussed my general health and wellbeing. I confessed to being tired, after a hard week the last time he saw me of preparing for a big client demo and then following it up the week after with energy-intensive marking. He wondered if I could afford to not do the work at the Open University. The financials do not allow that at the present time. Almost half of what my company paid me (pre-tax!) went out as rent here in England. Without the OU income, I would have been in the hole. Nevertheless, I did discuss with him my idea of working more at the OU, maybe up to three days a week and none for my own company. I would probably make more money, be less stressed, and have more time to spend on my Ph.D. Doing so, particularly without planning and warning, would place my own company in a precarious position, so I will need to muddle along as best as possible for the moment.
Next meeting is April 2nd at 16:30.
Task list:
- Resurrect and review Ph.D. plan.
- Sort out MATLAB issues
- Work on CSR version of my paper so that I can submit it for publication and send it off to my external advisor

Share/Save

Nevertheless, I still had to confess that I basically had not accomplished anything since our previous meeting on the 23rd of September and that our librarian was hounding me to submit my revised paper as I promised her I was going to publish it as a Computer Science Report last year. She now has a gap in her numbering and wants me to cough up the goods. My supervisor thought that would be a good place to start in getting my mind back into my research. Revision of that paper and publication of it is also what I have been waiting on before seeking assistance from an external advisor, so it all ties together nicely.
He also thought I should make sure that I am represented in the upcoming annual HCT publication, with a picture and a description of my project. As that was already on my list of things to do, that is not a big deal. I believe I can even recycle previous material. I showed him my to-do list and my latest calendar for working on my Ph.D., which reminded him that I should unearth my research plan and update it for past progress and new timelines. I am not looking forward to that, but it is good to have a timeline. My new proposed timetable gives me between 5 to 7 research days at the University a month.
I mentioned, almost in passing, that I will be winning a national award for excellence in teaching from the Open University. He was extremely pleased and suggested that I drop a note to the editors of the Bulletin, a weekly newsletter here at the University, as that’s the kind of tidbit they’d likely want to publish. That also led into a discussion of what I am doing at the Open University and my usage of FirstClass there. Our Lab is getting a FirstClass conferencing system for sharing resources and updating web pages. I wasn’t sure how well it would work for updating web pages automatically, as I have not seen FirstClass really used for that, beyond shared calendars, but I think it’s a neat experiment.
Finally, I told him that I’d actually put together a research blog which was currently very short on content and unindexed by Google, but it was a good place to start my efforts in building myself a community. It’s another positive step forward.
At the end of the meeting, we closed, as usual, by selecting a new meeting date. Next meeting is a half hour squeezed inbetween other appointments on Tuesday, the 16th of March, at 16:15.
Task List:
- Send another letter to postgraduate dean about extending intermission from October to end of February.
- Send HCT Info to Rose’s secretary for inclusion in annual publication.
- Resurrect and review timelines for my research plan.
- Send note to A.F. of the bulletin about winning teaching award at the Open University.
- Review the paper for submission to the library.
- Send e-mail to F.S. saying I’m alive and include revised version of paper.
- Review MATLAB application acquisition status.

Share/Save

This is my current revised program:
% Import anal1ahyp.txt into anal1a matrix variable. Do this first! It’s not done
% automatically here. Need first to do something like anal1a = abs(100 – anal1a) to
% get proper dissimilarity values to work with. The default ones in the files don’t
% work as is.
docs = {‘g4c’, ‘pp1′, ‘pp2′, ‘msc’, ‘pl1′, ‘pl2′, ‘pl3′, ‘sp1′, ‘sp2′, ‘ac1′, ‘ac2′, ‘bws’};
[Anal1aMDS, eigvals] = cmdscale(Anal1aRaw);
plot(1:length(eigvals),eigvals,’bo-’);
graph2d.constantline(0,’LineStyle’,':’,'Color’,[.7 .7 .7]);
axis([1,length(eigvals),min(eigvals),max(eigvals)*1.1]);
xlabel(‘Eigenvalue number’);
ylabel(‘Eigenvalue’);
plot(Anal1aMDS(:,1),Anal1aMDS(:,2),’bx’);
axis(max(max(abs(Anal1aMDS))) * [-1.1,1.1,-1.1,1.1]); axis(‘square’);
text(Anal1aMDS(:,1),Anal1aMDS(:,2),docs,’HorizontalAlignment’,'left’);
hx = graph2d.constantline(0,’LineStyle’,'-’,'Color’,[.7 .7 .7]);
hx = changedependvar(hx,’x');
hy = graph2d.constantline(0,’LineStyle’,'-’,'Color’,[.7 .7 .7]);
hy = changedependvar(hy,’y');
% Import anal2ahyp.txt into anal2a matrix variable. Do this first! It’s not done
% automatically here.
Anal2aRaw = abs(100 – Anal2aRaw);
[Anal2aMDS, eigvals] = cmdscale(Anal2aRaw);
% do procustes
[D,Z] = procrustes(Anal1aMDS, Anal2aMDS(:,1:2));
plot(Anal1aMDS(:,1), Anal1aMDS(:,2), ‘bo’, Z(:,1), Z(:,2), ‘rd’);
text(Anal1aMDS(:,1)+0.5,Anal1aMDS(:,2), docs, ‘Color’, ‘b’);
text(Z(:,1)+0.5,Z(:,2),docs, ‘Color’, ‘r’);
xlabel(‘East of the Sun’);
ylabel(‘West of the Moon’);
legend({‘Anal1a’, ‘Anal2a’}, 4);

Share/Save

The MATLAB discussion led quite nicely into the progress. I was able to show him a graph and explain what I had been doing and my current problems. He’s going to look for info on MATLAB courses at the university and/or tutorials so I can get up to speed faster. He also reminded me that David Young is a MATLAB resource I might be able to tap.
He was quite pleased about my progress and thought I’d managed to get quite a bit done, even if I hadn’t yet completely solved my analysis problem. At least I was getting somewhere.
Things To Do:
Clean up report and give to Ben before publishing as CSRP
Continue work on MATLAB
Read my Hypertext ’03 papers
Research what I was waiting for David and Ian to provide
Continue on with MATLAB licensing issues

Share/Save