Exploring Art, Theater and Sentence Entropy
with Machine Learning
Hello!
I’m Miles Thompson
Mostly I’m a data scientist and developer but sometimes I like to pretend I’m an artist ;-)
You can find me at miles@goodbright.nz
The great outdoors
Some Motivation
Table of Contents
◉ Context ◉ Data collection
◉ Entropy ◉ Other experiments
Computers
Machine Learning
About the director,
Annie Dorsen
Broadway – “Passing Strange”
Algorithmic Theater
- “Hello Hi There”
- “Yesterday Tomorrow”
- “The Great Outdoors”
“The audience will gather around the flickering projector, with all kinds of text and stories summoned down to us from the “cloud.” .. We gather together to hear the stories not so much from our ancestors as from each other, right now, two seconds ago.”
“
Development process
Gear 1
Gear 2
Some of the people involved
◉
◉ Director
◉ Music programming
◉ Video programming
◉ Text programming
◉ Dramaturgy
◉ Our amazing actor/s
◉
The theater process
◉ Theater - turns out “it’s a whole thing” ;-)
◉ Need very rapid iterations
( < 1 hour)
◉ Solution: Split code between preparation and script.
Adjust “script” code during rapid iteration, do numeric processing offline.
Data Collection.
Pushshift.io
Spark / Databricks
Power but $$
Motorway
more home-build but free
DEMO
A quick dive into some of the text
Entropy
The Arc of the piece
Lowest entropy
Lowish entropy
Middle entropy / More personal, long stories
Higher entropy
Even higher entropy
Very highest entropy
Shannon’s definition of entropy
◉.. minimum number of bits needed to encode
a string of symbols, based on the frequency of the symbols ..
◉ “ The form of H will be recognized as that of
entropy as defined in certain formulations of statistical mechanics where p i is the probability of a system being in cell i of its phase space.” -Bell System Technical Journal, 1948 “A Mathematical Theory of Communication“ Shannon, C.E.
Shannon’s definition of entropy
◉.. minimum number of bits needed to encode
a string of symbols, based on the frequency of the symbols ..
◉ “ The form of H will be recognized as that of
entropy as defined in certain formulations of statistical mechanics where p i is the probability of a system being in cell i of its phase space.” - Bell System Technical Journal, 1948 “A Mathematical Theory of Communication“ Shannon, C.E.
Quantifying a word's information content
entropy: how uncertain we are about what is being communicated at a certain point (it decreases with each upcoming word)
POS
INTERP
SIBLE RETATIONS
maximal uncertainty: all interpretations are equally probable
we are certain about one particular interpretation （ $ P ( x )=1 $ ）
after the first w
Shannon Entropy, very bad entropy and other measures
◉ Just counting how common a sentence is
◉ Shannon entropy
◉ Compression entropy (online compression)
◉ “Very bad entropy” (PySmaz)
Semantic‘Entropy – RNN Entropy
Semantic‘Entropy – W2V Entropy’
Sentence online with highest W2V Entropy?
James Joyce – Ulysses-for the win
Other experiments
Content Shaping
Semantic Resonance
Rhyming
On working with artists
Working with artists
◉ Don’t Dunning Kruger it
- realize how much you don’t know
◉ Don’t show algorithms without being
cognizant of other aesthetics – wait a bit more till ready?
◉ “Open source” ?
- can make some artists twitch
◉ Respect
- is spelled R E S P E C T
.. we live in a world in which the question of agency is highly disputed, in which our access to choice is circumscribed by the framing of political discourse, by the simultaneous cornucopia and contraction of “options” in our consumer paradise, and, indeed, by algorithms, which filter, consolidate and display certain possibilities while rendering all others invisible.
- Annie Dorsen, 2012
“
Thanks!
Any questions?
You can find me at
◉
@utunga
miles@goodbright.nz