That Rugged Raw

It has been a more-busy-than-usual week on campus, and I've had a pretty packed conference schedule. I've been much too far behind on New Year's Resolution:

Ship Copy.

So, in it's most-purest most-rawest most-honest form, you can find a number of raw transcripts in the new repo: decause/raw

So, What does that look like?

.
├── libreplanet
│   └── 2015
│       ├── closingkeynote-libreplanet-karensandler.txt
│       ├── debnicholson-friday.txt
│       ├── freesoftwareawards.txt
│       └── highpriorityprojs-libreplanet-friday.txt
├── pycon
│   └── 2015
│       ├── ninapresofeedback.txt
│       ├── pycon-day2-keynotes.txt
│       ├── pyconedusummit2015.txt
│       └── scherer-pycon-ansible-day2.txt
└── RIT
    └── 2015
        ├── biella-astra-raw.txt
        ├── biella-molly-guest-lecture.txt
        └── molly-sauter-where-is-the-digital-street.txt

6 directories, 13 files

$ wc -w libreplanet/2015/*.txt
 2287 libreplanet/2015/closingkeynote-libreplanet-karensandler.txt
  139 libreplanet/2015/debnicholson-friday.txt
  586 libreplanet/2015/freesoftwareawards.txt
 2297 libreplanet/2015/highpriorityprojs-libreplanet-friday.txt
5309 total

$ wc -w pycon/2015/*.txt
  106 pycon/2015/ninapresofeedback.txt
 2233 pycon/2015/pycon-day2-keynotes.txt
 1844 pycon/2015/pyconedusummit2015.txt
   63 pycon/2015/scherer-pycon-ansible-day2.txt
4246 total

$ wc -w RIT/2015/*.txt
 4521 RIT/2015/biella-astra-raw.txt
 2489 RIT/2015/biella-molly-guest-lecture.txt
 1964 RIT/2015/molly-sauter-where-is-the-digital-street.txt
8974 total

18529 total total

18,529 words, or, just over 41 pages total of raw text.

There is a flaw in my workflow. Though there is some utility in a raw transcript, really it is mostly when delivered in real-time. After the fact, there is much post-production work to be done, like spell checking. Even after, if there is a video, then the transcript is partial, and incomplete. This is bothersome to many potential downstream consumers of raw text. So where does that leave us?

Word Clouds

I've played with word_cloud before within my decause/presignaug for building presidential inauguration visualizations last year. Since then, word_cloud has gotten much more sophisticated--now using scikitlearn, and numpy, and providing the ability to fit word clouds within images!

List of Issues/Fixes

  • you'll need to pip install cython first
  • You'll need to sudo yum install freetype-devel (probably not necessary, since this is alleviated by pointing at a diff .ttf typeface...)
  • you'll have to edit your FONT_PATH within word_cloud.py
  • image masks *must* be saved as greyscale, not rgb images (this was a biggie, and I wouldn't have figured it out if GIMP didn't display the color encoding in the file statusbar when you opened things :) )
I went ahead and uploaded the changes I made to my fork on GitHub: http://github.com/decause/word_cloud if you'd like to see them. The important files are pycon.py and pycon-greys.py.

Comments

Comments powered by Disqus