Seq2Seq_ChatBot

Keras Flavour

I'm assuming the Theano-backend version of Keras. Make sure you have installed Keras.

Apart from this you'll need the following python libraries.

Garbage Collector : gc (already installed)

Numpy : sudo -E pip install numpy

Pickle : pickle (aread installed)

Usage

From the root of the github directory,

$> THEANO_FLAGS=mode=FAST_RUN,device=gpu,floatX=float32 python train.py

IMP Parameters

Inside ~/train.py, there are a set of global-variables,

#------- Global Variables ---------#
content_filepath   = 'new_data/out_TheSimpsons.tsv'
chars = '0123456789+/-*=abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ!.,?(){}[]&#_ ' # Char-Level Vocabulary.
TRUNCATE_SIZE = 20000
#----------------------------------#

Update these according to your needs (read infrastructure limits). There are other model-definition-parameters in the same script but it might not be the best of ideas to change them randomly. Feel free to try out though!

#-------- Parameters for the model and dataset--------#
questions = dataset.contexts[:TRUNCATE_SIZE]
expected = dataset.responses[:TRUNCATE_SIZE]

RNN = recurrent.LSTM
HIDDEN_SIZE = 512
BATCH_SIZE = 10
LAYERS = 3
X_MAXLEN = len(max(questions, key=len))
Y_MAXLEN = len(max(expected, key=len))

ctable = CharacterTable(chars, X_MAXLEN)
#-----------------------------------------------------#