blog.infochimps.org - Organizing Huge Information Sources

Do you work with huge datasets?  Are you interested in the opportunities that abound when the world’s free open data are integrated and the Semantic Web becomes a reality?  Come discuss data mashups, visualizations, and tools to organize, discover and explore rich information streams.

Join us at Mangia pizza on Guadalupe (3016 Guadalupe #100: free wifi, beer & great pizza) Thursday July 10th at 6:30pm and meet your fellow Austin data nerds.

(Also at Upcoming)

Rejoice, you open-source orangutans, for the powerful, the weighty, the Infinite Monkeywrench is now hosted on GitHub! Download a copy and start hacking, if you will, and send us your questions and concerns.

The Infinite Monkeywrench (IMW) turns all the screws in the heaving contraption we call infochimps.org but can also be put to good use on more modest projects as well. Learn more about IMW at the official IMW website.

Hooray! The infochimps have been waxy’ed.  Let’s see how the server bonobos stand up.

It’s been suggested that I highlight some of the “gems” of our collection, which we’re going to spend the whole weekend shoveling into the pile. These first few are really deep, and somewhat hard to get / not widely known:

• Full game state for every play of every baseball game in 2007, majors and minors.  Additionally, for about half of the major league games, *pitch by pitch* trajectory and game state information.  (MLB Gameday)
• Word frequencies in written text for ~800,000 word tokens (British National Corpus)
• All the wikipedia infoboxes, turned on their side and put into a table for each infobox type.
• 250,000+ Materials Safety Data sheets - the chemical and safety information required by OHSA
  
• 100 years of Hourly weather data; from 1973 on there’s about 10,000 stations all taking hourly readings … put another way, it’s 475,000+ station-years of hourly readings and weighs in at ~15 GB compressed.

(Incidentally, many of those datasets sell for inexcusable and malicious prices.  For those with a commercial bent, something tells me there’s room in the market if you’re willing to accept a markup of less than 10,000 times).

These are a bit silly but interesting for their ridiculous depth:
* A variety of mathematical constants (pi, e, Catalan’s number, the Golden Ratio, others) calculated to in some cases a preposterous 100 billion decimal places (I’ll probably chop them off at a still-ludicrous 500 million).
* 5000 years of solar eclipse times, 6000 years of precise lunar phase, 6000 years of venus transits.
* Odds of Dying for every Cause of Death listed in the US in a given year.

There are also, of course, the well-known collections: IMDB.com, musicbrainz, dbpedia, CIA factbook, geonames, citeseer, census, statistical abstract and the like.  So let’s see how much of the low-hanging fruit we can toss up there this weekend (the hard parts are adding metadata, and getting the non-copyrightable data out of the copyrighted screenscrapes, so what you’ll see are minimal metadata and the non-screenscraped datasets — still beats paying $1200+/GB though.)

[edit: dates for holidays by country, year-by-year odds of dying for all causes of death from the recent 8 year, NIST values for physical and chemical constants, mechanical properties of common engineering materials, and the spoken and written word frequencies for ~800,000 word tokens datasets should be up later today -- if the site is down briefly we're pushing that update to the server.  (If the site is down not-briefly we've been del.waxyslashdiggdotted)  Thanks to my friend Ned for helping do some drudge work to get those out.]

FINALLY — got the wikipedia infoboxen posted to the site, along with some tiny fixes.

This is 3000+ tables on everything from ABA Teams through Simpsons Episodes to Zodiac Signs.  There’s a fair amount of cruft in these, but until I have live metadata editing going I’m not going to worry about it: it takes about 8 hours start to finish to process this dataset, they’re not perfect but they are perfectly usable.

I have the weather dataset and baseball datasets almost ready to go (along with a whole buncha others), but I’m going to take some time to get the site running better first.  Here’s a rough TODO list:

1. live, versioned metadata editing
2. uploading
3. Allow grouping of datasets by collection and add category tags
4. Make it so fields & contributors tie together.  (For complicated reasons, each dataset creates a new personal version of the field so you can’t actually walk from one “stock price” field to other datasets with that tag.

Then I’ll turn some intensive attention finally to the InfiniteMonkeywrench code.  We need better tools to wrangle these huge datasets into shape.

Many people don’t know how broad our rights to factual data actually are.  Unlike the mishegaas that reigns in copyright land, the world of data is largely open (and rightfully so).  To arrive at the age of ubiquitous information with a sound policy, however, we have to exercise those rights assertively, respectfully and prudently.

Let me start with the traditional IANAL and point out that if you take legal advice from a chimpanzee you deserve what you get. Instead, read iusmentis on database law and bitlaw on compilations and databases. (In which case you can probably skip the rest of this post.) (Also, the following only applies to the US, where the database laws are actually more liberal than elsewhere; I have no idea what the situation is outside the US)

In general, a comprehensive assemblage of facts cannot be copyrighted. Copyright only applies where there is creative content. A comprehensive list of cars and retail prices cannot be copyrighted; a comprehensive collection of reviews of those cars can be copyrighted. A list of all the musical albums released each year is data; the lyrics and music within them is creative. A list of word tokens sorted by artist, genre, release date and song length is data, and a list of the top-100 selling albums by year is data. This is the important Feist Publications v. Rural Telephone Service case:

“Facts, whether alone or as part of a compilation, are not original and therefore may not be copyrighted. A factual compilation is eligible for copyright if it features an original selection or arrangement of facts, but the copyright is limited to the particular selection or arrangement. In no event may copyright extend to the facts themselves.” — Sandra Day O’Connor for the Supreme Court

“A collections of facts are not copyrightable per se … A compilation, like any other work, is copyrightable only if it satisfies the originality requirement (”an original work of authorship”). Facts are never original, so the compilation author can claim originality, if at all, only in the way the facts are presented. The facts must be selected, coordinated, or arranged “in such a way” as to render the work as a whole original.” — Sandra Day O’Connor for the Supreme Court

A presentation of data can be creative — you can’t xerox the blue book and hand that out. However, a conversion of otherwise unrestricted data into your own creative presentation satisfies this restriction. So would a presentation (original or converted) that did not arise from a creative act — you couldn’t claim copyright on a .CSV file of some dataset.

Besides “presentation” and a couple edge cases (”hot news”, “selection and arrangement”), the main one to be aware of is “Terms of Service“. If you have to agree to terms of service that restrict the data, but you take it anyway, you can be guilty of trespass. My understanding there is that if you can a) access the site by robot (no person clicks anything) AND b) there is no robots.txt, they shouldn’t be able to sustain a claim that it’s a restricted resource.

I personally go by balancing two principles:

1. It’s our world, and we deserve access to the information that describes it.  Besides our legal rights, we have an even stronger moral claim to the chronicle of our collective story.  And we all stand to benefit: there have to be incentives to gather and organize data, but the modest benefits of making a data provider a lot richer don’t stand against the much larger marginal benefit of making the world a timy bit smarter.
2. Be a good neighbor.  A lot of work goes in to gathering, processing, verifying, distributing an interesting dataset.  If we infochimps run around ignoring people’s requests for modest usage conditions, we’ll have a bit extra of open data and a lot extra of pissed-off ex-kindred souls who feel like we stole their cake.  Inevitably, this will mean that people won’t put data online at all for public access.

The best approach is

• Scrupulously credit contributions, make clear that their efforts are recognized, and that we’ll link back to them for their ultimate benefit.
• Clearly state the usage restrictions requested by the contributor, adhere to them, and ask that recipients of the data do the same.
• Make clear the benefits to the world for making this data available.
• Make clear the benefits to the contributor — this data will, for free, be enhanced with metadata, converted for use by diverse tools, interlinked with other rich datasets, and power interesting projects.  If your mission statement is “build reliable and exciting cars” or “make powerful music”, then your mission statement isn’t “explore and explain unexpected correlations among disparate rich information pools”.  Let someone else do it for you, and let them build the tools to do so around your data.  Consider how much Baseball has benefitted from its statistical revolution — fed by its incredibly rich ecosystem of open data.
• Finally, as far as scientific or government prepared data that’s otherwise rights-free: gloves off, we’re taking that data.  If you’re a researcher, and you’re not openly sharing your data, you’re not only a bad scientist but also a bad person.  Ditto for data collected at taxpayer expense.

40 Years of data on every NYSE, AMEX and NASDAQ listed stock:

• AMEX Listed Stocks 1970-now, daily open-close-low-high-volume, as well as dividends.
• NASDAQ Listed Stocks 1970-now, daily open-close-low-high-volume, as well as dividends.
• NYSE Listed Stocks 1970-now, daily open-close-low-high-volume, as well as dividends.

These links were busted before but should be worky now.

Added the Statistical Abstract of the United States — the messily, messily formatted analyzed tables released by the US Census Department.  1350+ tables, yum.

Just in time for SxSWi - the site is live.

Now that we’ve got the skeleton of the website in place, we can go back and apply the necessary metadata/package/import workflow we’ve developed.

Here’s a rundown of the datasets you can look forward to seeing over the next few weeks:

• demographics
  • world
    • world bank development data—variety of country data from world bank
    • CIA factbook
  • us
    • Statistical Abstract of the US —an exhaustive categorization of demographic, commercial and social data for the US
    • The full US Census Summary File 3, at the zipcode level.

• money:
  • US Stock market daily—Daily open/close/lo/hi for all listed stocks since 1970
  • US Campaign finance—Expenditures in US presidential, senate, house and governor races in 2004
  • Constantcurrency—Variety of currencies in constant dollars/pounds etc back to the 1600s

• huge & Miscellaneous:
  • infoboxen:—All the infoboxes from Wikipedia broken out into individual semantically labelled tables

• joins:
  • Common coding systems for
    • country codes, including a useful keying database from common names (“USA”, “U.S.A”, “United States”, …) to ISO country code
    • languages
    • currencies, etc.
  • time—conversions among all of the (curiously many) competing means of measuring dates and times

• health
  • odds of dying—all causes of death in the US, broken down by category and given as rate and odds
  • middle east conflict casualties—civilian and military deaths in Iraq (OIF) and Afghanistan (OEF) since 2003/2001

• science, math & engineering:
  • nasa_eclipse 5000 years of solar and lunar eclipse, lunar phase, and planetary transits from NASA
  • 270,000+ MSDS (Materials Safety datasheets) listing properties and hazards of common and industrial chemical substances
  • material properties—basic chemical and physical properties for common chemical substances
  • powergrid Network of Power Grid Connections in the Western US (Strogatz199
  • fastenerdata Screw, Bolt, and Threaded Fasteners: Dimensions, Mechanical Strengths and Properties, and other useful information
  • mechanical properties Mechanical Properties for a variety of useful materials
  • consts and units Universal constants and unit conversion factors
  • standard mathematical tables Tables of Elementary functions (log, bessel, etc) over large range
  • mathematical constants The fundamental mathematical constants calculated to millions and occasionally billions of decimal places

• Art and Culture
  • Every movie, act(or|ess), and film courtesy of imdb.com
  • Every musician, album, track and label, courtesy of musicbrainz.org
  • WANTED:ISBN=> author, book, publisher dataset. If you have this please contact us.

• geo:
  • A huge assortment of GIS layers from nationalatlas.gov
  • Geographical place names & locations from geonames.org
  • TigerLine, a mapping from street address to location for the full US (this will take a while)
  • Postal codes – map from zip code to city and latitude/longitude

• time
  • tzinfo time zone info for everywhere
  • calendar_kitchensink 3000 years of time zone, calendar conversion, moon phase, accounting information, etc
  • accounting_calendar last fridays of each month, adjusted for holidays etc.
  • holidays major repeating holidays for most countries

• language:
  • Usage frequency (in speech and print) of every english word, from the British National Corpus
  • Moby Word lists – Word Lists, Multiple Language Lists of Common Words, Hyphenation, Part of Speech, Pronunciation, Thesaurus
  • Natural Language Toolkit Corpora NLTK’s Word lists, semantic networks, lexical data, large text corpora; several languages
  • All the words legal to play in Scrabble™

• sport:
  • Baseball:
  • retrosheet gamelogs: Game outcome and box score for every MLB game back to 1890s
  • retrosheet event files: Play by play information for almost every game back to 1957 (and all since the mid-1970s).
  • baseballdatabank: Season and Career stats for every MLB player, team, etc of all time
  • MLB Gameday: Players, Game state, Pitch-by-Pitch trajectory and Outcome for ~half of the 2007 MLB games.