alignment

KO | EN

gitlite — search

#hacktoberfest #react #ai #go #machine-learning #deep-learning #linux #macos #nodejs #awesome #cli #pytorch #security #open-source

(★ 8)

Conversational alignment

alignment Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# alignment
Conversational alignment

Code examples
    - Reading in a csv
        https://docs.python.org/2/library/csv.html
    - Grouping list of dictionaries
        https://www.daniweb.com/software-development/python/code/216750/group-a-list-of-dictionaries-python

Formulas
    - TRUE_POWER (Outputs the expected results)
    - DNM (Formula used in [Echoes of Power](http://www.cs.cornell.edu/~cristian/Echoes_of_power_files/echoes_of_power.pdf)

## changelog

**6/26/2015**

Derived formula for Echoes of Power:
    powerProb = ((# of times marker is said by A and B)*(# of words A says to B))/((# of words B says to A)*(# of times A says the marker))
    baseProb = (# of times B says the marker)/(# of words B says)
    prob = powerProb - baseProb

**6/29/2015**

Figured out that the formula we were working off of was incorrect.

Instead of calculating probabilities using tokens, we need to calculate probabilites using utterances
    powerProb = (# of utterances where A and B both say marker)/(Number of utterances where A says the marker)
    baseProb = (# of utterances where B says the marker)/(# of utterances that B says to A)
    prob = powerProb - baseProb

**7/13/2015**

Figured out that the formula we were working off of was incorrect.

Our new formula is:

powerProb = log((# of utterances where A and B both say marker)/(Number of utterances where A says the marker))
baseProb = log((# of utterances where B says the marker and A doesn't)/(# of utterances that A doesn't say the marker))
prob = powerProb - baseProb

**7/20/2015**

SHARED_CODE:
    Fixed a bug which was reducing our BNOTA

**7/22/2015**

SHARED_CODE:
    Added DNM formula option and options to change smoothing values

**7/23/2015**

TWTR:
    Updated readCSV to discard more utterances which quote a speaker

- Check for unicode quotes
    - Check if speakers acknowledge each other
    - Check if utterance contains "RT"
    - Check if utterance contains "[mention]: "

**7/27/2015**

- No longer discard if speakers don't acknowledge each other

alignment

alignment Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

alignment

alignment Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README