cherylyli / char-rnn

ObamaBot for qHacks

#ObamaBot Tweet at ObamaBot at @PseudoObamaBot! For best responses tweet political topics.

I used a recurrent neural net with 2 hidden layers, a dropout rate of 0.5, and rnn size of 521. A 1.5MB text file of Obama speeches to generate Obama-Style speeches. Then I used a Markov Twitter bot to read from the output of the Obama-Style speeches and generate short tweets with certain keywords that Donald Trump supporters tweet out.

Used a bunch of pre-existing libraries. Notably:

• char-rnn for the recurrent neural net
• using pseudoObama-Style text as input to respond to keywords extracted from Donald Trump's tweets
• indico to parse keywords and intent of tweets

Install a bunch of things, I recommed using a separate environment running python 2

• download torch / Lua
• run this command, this will take a pretty long time to run though.

$ th train.lua -data_dir data/Obama -rnn_size 512 -num_layers 2 -dropout 0.5

• get some output Obama gibberish: (you can add a length and primetext and temperature, these are fun things to play with)

$ th sample.lua cv/some_checkpoint.t7 -gpuid -1 

• you can redirect your ouput of Obama gibership into a text file and let the markov model read from it
• set up twitter app -run

python ObamaBot/app.py

Used the following open source libraries:

• char-rnn
• MarvokBot

License

MIT