This repository contains the source code for the models used in the following paper:

The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems arXiv:1506.08909.

• Python 2.7
• Theano bleeding-edge
• Lasagne (specifically, this fork with commit a3890b2a743e7341c337e73a133120fbebee4150). This will be cleaned up when recurrent layers are merged into Lasagne.
• Pyprind

Fetch the pickled data:

cd src
wget http://cs.mcgill.ca/~npow1/data/ubuntu_blobs.tgz
tar zxvf blobs.tgz

Note that this code has been heavily modified to support many different models. To reproduce the results in the original paper, use the following incantations.

RNN:

python main.py --encoder rnn --batch_size=512 --hidden_size=50 --optimizer adam --lr 0.001 --fine_tune_W=True --fine_tune_M=True --input_dir dataset_1MM

LSTM:

python main.py --encoder lstm --batch_size=256 --hidden_size=300 --optimizer adam --lr 0.001 --fine_tune_W=True --fine_tune_M=True --input_dir dataset_1MM

TFIDF:

python tfidf.py