Wednesday, May 31, 2006

O-Matrix version 6

Harmonic Software have released version 6 of their Matlab-clone software O-Matrix. The new release adds some basic statistics functions and some performance improvements.

The statistics are mostly so basic (random number generation over different distributions, a small number of PDF and CDF functions, descriptors like mean, median, covariance etc) that long time users, have no doubt added them themselves.

The performance improvements are more impressive when compared to the previous version but don't seem out of line with other software on my machine. I have not done a proper comparison, and the comparison provided by Harmonic, rather misleadingly compares this new version to Matlab 7.01 (released in 2004). Either they don't have Matlab 2006a or that gives an unfavorable comparison!

But this is all a distraction from a more fundamental issue- O-Matrix is still fundamentally a numeric-only matrix manipulation system. In the 80's, when the alternative was gluing FORTRAN libraries together, this was a big step forward. But now there are much better simulation languages. Of course the same is true of Matlab. But Mathworks, I think, recognizes this and puts comparatively little effort into Matlab development. Their value is in the toolboxes and Simulink- Matlab is a just a rather dated and mundane component. Harmonic have little extra to add to O-Matrix.

The only Matlab customers they can attract with a faster cheaper version of Matlab, are those that think that the world begins and ends with matrix manipulation.

Friday, May 26, 2006

Is the new TI Calculator good news for TI?

Texas instruments are about to launch a new model of handheld calculator, with the rather long name of the TI-Nspire™ CAS+.

At first glance this is an obvious and beneficial evolution of their product line. It appears to have a much higher resolution grayscale display like basic Palm's or Pocket PCs, and they have used this opportunity to tweak the software so that instead of having to switch between a spreadsheet view, calculator view or graph view, they can be displayed at once in a basic windowing system. It can also drive a full size monitor.

But I think that TI may be being drawn into dangerous ground. Consider the strength of their position, which as I commented at the start of the year, is based on calculators being supported in the classroom and exams but computers not. As the TI Calculators become more like handheld computers the pressure will increase to allow a wider range of computer solutions. As the conceptual gulf closes, TI's software (barely changed Derive from 15 years ago) will come under closer scrutiny. In the calculator world it is very impressive, in the PC world it would lose to almost every technical system available on capabilities and most of them on ease of use.

It is also only a matter of time before pocket PC's become routinely powerful enough to run PC based scientific software, blurring the distinction further(and perhaps only a few more years before your mobile phone can).

But what can TI do? They can't resist the pressure to advance towards this state, like lemmings drawn to the security of the lemming in front. Surely the solution is some serious investment in their software, but they haven't shown any interest in that yet.

Update 7 Sep 06: TI Nspire Calculator delayed
Update 23 Nov 06: Derive to be discontinued

Tuesday, May 23, 2006

Comparison of math typesetting software - Part 2

Continuing last weeks examination of math typesetting software, following are some comments on the capabilities of the systems.

This is a big subject, so picked a few choice features to look at closely:
1) As examples of non-trivial expressions I looked at two nested constructs: how does sqrt(sqrt(sqrt(.... look, and how does 2^2^2^2^2 look?
2) Do the systems have any automatic layout? IE What happens if you can't figure out how to fit an expression on the page, what happens if you re-size the page or fonts so that it goes off the page?
3) Antialiasing. These days its embarrassing to give a presentation by printing on transparency and using an OHP, so expressions should look good on screen too.
4) Detail control. What can you do, if you don't like the default choices?

Maple:
Maple was the only system that made no attempt to handle the nested expression test in any special way. Because the sqrt sign gets bigger as it gets nested, the leading slope, which is attractive for a small sign, takes up a lot of room as the sign gets bigger, making this expression very wide. All the other systems used a steeper, more compact, sqrt sign for the outer placements. Likewise, there is a drop in font size to the first superscript, but none for the subsequent superscripts.
There is a layout control system, typing a very long expression next to the pi in the test caused it to line break at the page width without breaking the meaning of the expression. sqrt signs were replaced with new parenthesis and ^1/2. However, it rather unnecessarily chose to put a line break after the opening parenthesis each time, resulting in lots of unused space. Oddly the closing parenthesis all happily sit on the same line.
Maple managed to antialiase the hard parts just fine (parentheses even when they are stretched to accommodate large expressions, sqrt signs etc), but didn't antialias any characters, which one might expect to be provided by the OS for free. (Click on the image to see the full size version, to see the jagged pi and 2s).
There appears to be no control over the look of typeset expressions beyond font characteristics (size, font, color etc).
Overall pretty basic, considering the system has just had a major overhall.

MathType:
MathType does a little better on the nested test. First and second sqrt are different and first and second superscript are different, though third occurrences do not change.
There is no automatic layout in MathType, make the expression too large for the page and it just gets wider and wider, as if you are on an infinite page, so it is up to you to plan the layout of a large expression. Without a page preview or any indication of the page width on the ruler, you may well waste some paper.
Antialiasing is the way round I expected. Characters handled nicely, sqrt signs showing jagged edges. Spanning parenthesis are handled well.
There is some basic detail control s- you can set the relative size of certain fonts (eg the superscript is by default 58% of the size of the parent character), but not many of them.

Mathematica/Publicon:
Mathemetica performs well on the nested test. There are five levels of different sqrt sign, and nest superscripts keep getting smaller indefinitely until a floor font size is reached.
Automatic layout happens in two ways. Expressions break automatically as the page width is exceeded. In this example, Mathematica switches to a prefix version of a sqrt sign and parentheses, instead of the ()^1/2 favoured by Maple. You can also ask it to do an automatic layout to try and fit the expression the best way that it can, though this discards any character level settings and may replace linear notations like ^ or / that you have chosen with superscripts and fractions.
Mathematica fully antialiases.
One difference is that math symbols are provided in two new fonts, which switch depending on whether the surrounding font is mono-spacing like Courier or proportional like Times. Whether you think this is a good thing is a matter of taste, but you can use the Symbol font like the other systems, if you prefer.
The place where it seems to stand out is in detail control. There is an overwhelming number of options from how thick the horizontal line is in a fraction, to how high or far across the nth root value appears in a radical. There are perhaps a couple of hundred options and they can be set for a single character, the document or as default.

Scientific Word/Scientific Workplace:
Scientif Word came second in the nesting test. There are three levels of sqrt, though it makes these go further by using the third at the fourth level of the test, and two levels of superscript size.
There is no layout control, suffering from the same infinite page as MathType, though it does have a page preview.
Only characters are antialiased, and with the quite compact sqrt sign, this is probably the I think the worst screen representation from my examples.
Detail control was good, with access to lots of TeX tags, though the fact that these are applied through named tags was very frustrating. One doesn't appear to be able to select a character and make it 48 point. You have to edit a tag to have a 48 point and then apply the tag, being aware that any other use of that tag will inherit that value. Style sheets and abstraction like this are good, and ScientificWord provided a good range of pre-defined style sheets and tags. But to be forced to work like this is frustrating. I don't know if this is imposed by the use of TeX as the underlying language or if it is Mackitchin imposing good practice on us!


All systems claim to export into HTML, MathML, and TeX. I have not looked closely at this. Scientific Word and Publicon also support a range of target specific TeX or XML formats (eg specific journals).

So in conclusion, the Mathematica family performed best under these criterion, followed by the Scientific Word family. Though I repeat, there is a lot more to this subject, and this review was just a sampling of the feature space.

Putting this with my experience in the first part of the review, the Mathematica family are the clear winners (Mathematica if you have $1000+ and need computation, or Publicon if you have $150 and don't). For simple and easy to learn I still like MathType, though I was rather surprized by how basic it is considering its long pedigree and well known name. But at $130 its not expensive.

Friday, May 12, 2006

Silicon Graphics files for bankruptcy

Once the leader in high end visualization, Silicon Graphics is now on its knees. Court approval for its bankrupcy status will keep SGI going for a while: "We are pleased with the approval of our 'first-day motions' by the Bankruptcy Court," said Dennis McKenna, chairman and chief executive officer. "This approval will enable SGI to operate globally and meet normal business obligations."

He describes plans to reconstruct the business having shed most of its debt in a deal that sees existing shareholders lose their remaining share value. But any remaining confidence in the long term adoption of SGI technology must now be gone, so it is hard to imagine him turning it around.

It's sad. Ten years ago I was given a tour of SGI offices in Switzerland by a friend and it seemed the greatest place work - superb offices, excited staff and everyone had the most amazing (for the time) SGI workstations on their desks.

But a decade is a long time in IT.

Thursday, May 11, 2006

Comparison of math typesetting software - Part 1

Since I haven't done much real work on this site for a couple of weeks, and "comparison" is the keyword that brings a lot of people onto the site, I took a close look at the typesetting features of several products. My review covers: Maple, MathType, Mathematica/Publicon, OpenOffice and Scientific Word/Workplace.

Scope:While mathematical typesetting is only one part of publising technical papers, this review compares only that aspect. Some of the products do much more than this (Maple, Mathematica, Workplace also offer computation, data analysis, graphics etc, and Publicon, Scientific Word,/Workplace handle things like bibliographies, references, indexes etc), but that is not covered here.

The comparison will come in parts and today I will start with input. How do you actually enter your equation.


The first thing you notice is that all but OpenOffice have essentially similar kinds of palettes or toolbars, organized by types of symbols or structures. eg a palette for greek letters, a palette for matrix input etc. I was somewhat shocked by the absence of this from OpenOffice, which requires a language based input, based loosely on TeX type conventions. While keyboard input is important (and varied much more between products), I found the complete lack of mouse input frustrating as I learned the system and dropped OpenOffice from the rest of the comparison.

Maple:The palette system was neatly organized in a single column format and was easy to use, though you have to tidy as you go, as opening more than one section, quickly pushes parts off the screen.

You cannot type a keyboard combination for any special characters during typesetting, though you can type an entirely text based version of the whole expression and then format it.

Basic structures like superscripts and fractions happen automatically as you type the 1D version (^,/) which initially seems appealing, but means that you cannot type an actual "^" or "/" so expressions like appear to be impossible. Instead you must settle for the somewhat less clear expression on the right.

There doesn't seem to be any way to customize input, either with keyboard shortcuts or new palettes.

Maple does have the unique handwriting recognition feature, but regular readers will know that I was not impressed with that.

MathType:The surprisingly small to download MathType (4Mb), is described as the "Professional version of Microsoft Word's Equation editor". It has a very intuitive palette layout, though it was a quite sluggish to respond, and since double-clicking on a button brings up a button-properties editor, I found it frustrating to enter nested constructs like Sqrt(Sqrt(..

However, apart from a keystroke CTRL-Shift-G to go into a greek character mode, none of the special characters or constructs had built in keystrokes, though you can customize most palette buttons with a keystroke of your choice. [[Correction: Bob Mathews of Design Science points out that most have keystrokes, which one can see in the status bar on mouse over. I took a while to look at these, and it is true, perhaps 80% of the characters have preset keystrokes. Many are rather cryptic (ctrl+shift+k then # to type an Or symbol) making it hard to imagine knowing many of them, but the ability to change them mitigates this]]. There was a nice array of more complex templates organized by category, and you can also create your own.

Mathematica/Publicon: Publicon (the no-computation version of the Mathematica typesetting system) has the more elegantly organized palettes, with the nice feature of being able to scroll through the different sections with the mouse wheel), the Mathematica incarnation had separate palettes for different types of input, which is probably preferable if you plenty of screen space.

Neither had keyboard shortcuts to the palettes, but all individual structure elements and characters have keyboard access. The approach is the most thorough of the systems that I looked at, with every character having both a full name and a short name. Full names are very consistent, eg "\[Infinity] \[Alpha] \[LeftRightArrow] "etc. Shortnames were sometimes well chosen, sometimes less well and delimited with the escape key. eg "esc => esc" for \[Implies]. New keyboard short names and palettes can be created for any typeset structure in Mathematica but not in Publicon.

Scientific Word: The palette system has a neat button type that lets you tear off submenus to make separate palettes, but was otherwise the least well designed. eg the Sqrt button only comes with two placeholders (for the nth root) so if you want the more common Sqrt you must delete one of the placeholders, and rather a lot of cryptic icons (an Italic E means "Emphasized", an "A" with an arrow pointing to a "B" then an arrow pointing up represents search and replace!). There are no high level compound templates, and while you can customize which characters appear on palettes, you cannot create templates. [[Correction 23 May, thanks to Fred Chapman] - Shortcuts exist for all constructs and keyboard entry is provided for all characters by holding down ctrl and typing the TeX name of the character and then releasing ctrl. This is a system that mostly works well, except where the name contains an upper case letter. Trying to press and release the shift key without releasing the ctrl key can lead to some contortions.] Judge for yourself from the screen shots, but I also thought the palette design was rather ugly - though that doesn't affect its usefulness, its nice to work in a nice place!

Overall, when I compare input methods, I preferred MathType for entry level and Mathematica/Publicon as I became more experienced. Maple comes next then Scientific Word/Workplace with OpenOffice in a clear last place.

Continue to Part 2 - Which actually creates the best output?

Wednesday, May 10, 2006

Roundup of non-news

There seems to be a lack of serious announcements at the moment, so here are some of the items I noticed which did not deserve serious comment...

In the category of "new marketing initiatives":
New bi-weekly MapleSoft podcast - have someone else read their website out to you.
Mathworks sponsors some student engineering competition (mostly with software).
Mathworks gets some noteworthy user to speak at their conference - interesting if you have heard of Thomas Scharnhorst.
Wolfram Research are giving away some Mathematica to China, and sending some people there on Mathematica roadshow.
Wolfram Research offer free online seminars to learn Mathematica.

In the "business as usual" category:
Insightful Q1 results out - 'steady as she goes'
MathSoft hire a new distributor - but that was before the agreement to sell MathSoft (oddly still not announced on the MathSoft site)
MapleSoft buys an algorithm- but they have always bought in external code, so what's new?

Still, at least those companies have something to say about themselves. Design Science, SciFace, UNISTAT and VSN haven't put a news item on their sites this year. Come on guys, surely you have done something worth mentioning in the last five months?

Friday, May 05, 2006

Do scientists and engineers prefer Firefox?

It's just anecdotal evidence, but I was reviewing the site statistics for this site, and one piece of data stood out as interesting.

The browser used to read the site is running, fairly consistently at about 70% Firefox, 20% Internet Explorer, and about 10% combined of Opera, Safari and (new to me) Galeon.

Now I know that this is not normal, Firefox penetration is claimed to be around 20%, and professional sites that I am involved with are not much different.

Of course, it's small data, (my site delivers only a couple of thousand pages per week), and the self selection is much more specific, so "people interested in scientific computing, who are interested in blog comment, and can find a site that has barely promoted itself - prefer Firefox"!

The rest was all what you might expect, the average reader is American, British or Canadian, reads two pages from the site, is 25% likely to return, 10% likely to stay for over an hour and 95% likely to have got here through Google.

Tuesday, May 02, 2006

Visual Numerics moves to Texas

In a strange and slightly rambling press release, Visual Numerics explains that it is moving its headquarters from California to Texas, along with some comments about corporate strategy that are so vague as to be almost meaningless.

There are some suggestions as to the benefits of Texas, from density of Fortune 500 companies, to the value of the oil industry and presence of universities. But no real coherence to the case or conviction in any of the points.

I think that this may just be a case of reducing office rent costs and hoping that if you talk enough, people will think you have said something interesting about why.