Socrates is a chat bot based on Seq2Seq learning.

You will need to download word2vec pre trained file from https://code.google.com/archive/p/word2vec/ and place it in the data folder. The following libraries are required

1. Theano
2. Keras
3. Seq2Seq
4. GenSim

Check the requirements.txt for full list

Please place conversation files in data folder. The expected format of conversation file is each conversation is a text file, and in that file each conversation is a separate line. See the sample conversation file in /data/dummy_convo.txt

Clone the repo and run socrates/socrates.py

Words are converted into vectors using word2vec. Each line is converted to a matrix of size (words_in_sentence, word2vec_dimension). If sentence is less than words_in_sentence then it is padded with 'EOL' which indicates end of line. The tensor created per conversation file is of the dimension (samples, words_in_sentence, word2vec_dimension).

The model tries to predict next line from the current line.

All the unit tests are in the file /socrates/Tests.py