Author Archives: stern

Memory conformity

There was a nice article in Science this week about Memory Conformity, the phenomenon whereby people will alter their own memories to conform with versions of events claimed by others. In the study, even people who correctly remembered a test event would persuade themselves that they had seen something different if their peers claimed to have seen it.

The study used brain imaging to try to figure out how this happens, and their results seem so vague as to be of little use. However the phenomenon itself is fascinating, and I wonder if it explains anomalies like the Angel of Mons, a fairy tale about angels appearing in the sky over British troops in WWI, subsequently believed by some of the people who were there.

The theory seems to me relevant to the financial markets, as the number of opinions in the market tend to collapse to a much narrower range than the breadth of experience and knowledge present would seem to justify. Most people abandon their own memory of precedents, and their own intuitive or formal models, and buy into the consensus narrative for why housing prices should be high, or internet stocks, or whatever.

A line of bicycles at work this morning

The building in which I work requires bicycles to use the freight elevator during business hours. I have been following this rule for several months. Today, I arrived later than usual and for the first time found a line.

There was a fixed gear, a beater, a Brompton folding bike, a hardtail mountain bike with slicks. None of it casual, though. These were all bikes that saw a lot of use.

Thought Pattern

I am among the millions of users of Evernote, and I have been struck by its many similarities to a software package from the early 1990s called Thought Pattern™. This was a note-taking package that made little impact on the market but which I found very useful at the time.

As with Evernote, a Thought Pattern user can enter notes with text and pictures, and assign keywords to each note. This makes it possible to search later by the terms in the card, or by themes that the cards have in common, even if the name of the theme appears nowhere in the text of the card itself. Unstructured data can be challenging to manage. With a large unstructured database used by many people, communal usage patterns can train the system about related content even in the absence of identical terms. However, for smaller datasets and user bases, explicit tagging of the sort used by Evernote and Thought Pattern is much better. Highly structured data can be handled in a traditional database management system, but a lot of information in the real world doesn’t conform to any simple rigid system of fields and records.

Thought Pattern also allowed the user to link content from other applications to cards, where it was represented by icons and would open if double-clicked. The program had one major feature that Evernote lacks — it was possible to attach an alarm to each card. Thought Pattern also had a better logo.

thought pattern logo

I rode a bicycle across most of the United States in 1993 wearing a Thought Pattern t-shirt.

Thought Pattern lacked of course the networked ubiquity that makes Evernote so useful. Nonetheless, it was a great program that I found useful. Credit to Stephen Zagerman of Bananafish Software for being ahead of his time.

in action

box back

Date functions in Mathematica

The date functions in Mathematica are in some regards excellent — you can use almost any common date format and the program will decipher it, and it computes date additions and date differences to high precision, and the date graphing functions work well. However these features come at a cost, the date functions in Mathematica are quite slow.

When writing functions that involve dates, the most important speed optimization is typically to move all date functions outside the inner loops, even if this means doing elaborate computations of all possible date combinations in advance, so that the inner loops can do table lookups for needed results. I have even created my own date functions that, though interpreted rather than compiled as Mathematica’s built in functions are, run ten times faster. (One difference being that my function is less flexible in the format of the inputs it accepts, and can proceed directly with date addition or subtraction, whereas Mathematica’s built-in function must spend epic amounts of time trying to figure out what precisely is meant by, for example “11/5/1980”. However even when you convert all Mathematica dates to AbsoluteTime[] format, it’s still slow.)

I ran some benchmarks tonight, comparing Mathematica to Visual Basic for Applications. This seems a fair comparison, as both are interpreted lanaguages. The inner loop in Mathematica looks like

{DateDifference["11/15/1997", "12/1/1998", "Day"],
DatePlus["11/15/1997", {36, "Day"}],
AbsoluteTime["December 30, 1997"]}

While the equivalent code in VBA is

diff = DateDiff("d", #11/15/97#, #12/1/98#)
sum = DateAdd("d", 36, #11/15/97#)
intTime = DateValue("December 30 1997")

Mathematica 8.01 runs that triad of lines 121 times/second, while VBA completes it about 10,000 times/second. VBA is about 82 times as fast.

An amusing quine in Mathematica

I had the idea for this Quine even before writing the longer, more traditional quine I posted yesterday. That one actually took hours to craft; this one practically wrote itself. It is based on the TextRecognize[] command that was introduced in Mathematica 8. Essentially, I wrote a line of code that prints a picture, then runs OCR on that picture and prints the result. The seed picture is an image of the program’s own source code.

Mathematica reduces the size of the picture so it fits neatly on a line of text, which makes it hard for a human to read. As they appear in the notebook, the Quine and its output look like
image quine

If we zoom in on the guinePNG image, we see


The only time consuming part of the process was selecting a font that led to an accurate OCR result. The TextRecognize command seems to do best with dictionary words, and does a comparatively lousy job of recognizing Mathematica code, which is full of odd terms and punctuation. I changed the name of the variable from “quinePNG” to “guinePNG” because TextRecognize[] was identifying the “q” as a “g” anyway.