Torch Neural Conversation Model

This is an implementation of seq2seq for language models implemented in torch.

The main features are:

1. Multilayer LSTM's
2. Batch Processing
3. Explicit Vocabulary Size
4. Adagrad (but easy to use any torch.optim plugins)
5. Train/Test split
6. Gradient clipping
7. Large dataset support (above the normal LUA JIT limits, but limited by your system RAM)
8. Beam Search for Decoding

This is really an extension of awesome work from the Element Research People: (rnn) and macournoyer's great project (neralconvo) and some helpful code from karpathy's char-rnn.

This is technically in beta form, but I have confirmed that it is working.

##Examples

I did a quick training with the first 9 million examples of the opensubs dataset for three epochs (with minibatch of size 1, adagrad learning rate .01, 25 words in, 25 words out)

These outputs are the top five beams (using th beam.lua)

Ask: hi

• Hi.
• How are you?
• What are you doing here?
• What are you doing?
• How are you doing?

Ask: where are you from ?

• I dont know.
• Im from ohio.
• From the north.
• I dont know...
• I dont know...... but i dont know.

Ask: how old are you?

• <number>.
• Im <number>.
• I dont know.
• <number>?
• <number>, <number>.

Ask: goodbye

• Goodbye
• Goodbye.
• What are you doing?
• Goodbye...
• What are you doing here?

##Installation

Tests

To run, use th trainoptim.lua --cuda

(well- you don't have to use cuda, but this would be crazy...)

When you want to test the model, run th beam.lua --cuda (or without the cuda flag if you trained it some other way)

##Dataset

I put a small sample from the opensubs dataset up. Really, you can add any dataset in the form:

input | response

with the pipe ('|') dividing the two. You should preprocess your data a bit if you use it like this. (Lua isn't the greatest for writing this kind pf preprocessing)

every new line is a new pair.