Skip to content

Commit

Permalink
Update README.md
Browse files Browse the repository at this point in the history
  • Loading branch information
@llSourcell
llSourcell committed Jun 5, 2016
1 parent 06b18b4 commit 86abfe8
Showing 1 changed file with 31 additions and 82 deletions.
113 changes: 31 additions & 82 deletions README.md
View file
Original file line number Diff line number Diff line change
@@ -1,23 +1,38 @@
# Neural Conversational Model in Torch
Overview
============
This is an attempt at implementing [Sequence to Sequence Learning with Neural Networks (seq2seq)](http://arxiv.org/abs/1409.3215) and reproducing the results in [A Neural Conversational Model](http://arxiv.org/abs/1506.05869) (aka the Google chatbot). The model is based on two [LSTM](https://en.wikipedia.org/wiki/Long_short-term_memory) layers. One for encoding the input sentence into a "thought vector", and another for decoding that vector into a response. This model is called Sequence-to-sequence or seq2seq. This the code for 'Build a Chatbot' on [Youtube](https://youtu.be/S_f2qV2_U00)

This is an attempt at implementing [Sequence to Sequence Learning with Neural Networks (seq2seq)](http://arxiv.org/abs/1409.3215) and reproducing the results in [A Neural Conversational Model](http://arxiv.org/abs/1506.05869) (aka the Google chatbot).
Dependencies
============

The Google chatbot paper [became famous](http://www.sciencealert.com/google-s-ai-bot-thinks-the-purpose-of-life-is-to-live-forever) after cleverly answering a few philosophical questions, such as:

> **Human:** What is the purpose of living?
> **Machine:** To live forever.
1. [Install Torch](http://torch.ch/docs/getting-started.html).
2. Install the following additional Lua libs:

## How it works
```sh
luarocks install nn
luarocks install rnn
luarocks install penlight
```

To train with CUDA install the latest CUDA drivers, toolkit and run:

The model is based on two [LSTM](https://en.wikipedia.org/wiki/Long_short-term_memory) layers. One for encoding the input sentence into a "thought vector", and another for decoding that vector into a response. This model is called Sequence-to-sequence or seq2seq.
```sh
luarocks install cutorch
luarocks install cunn
```

To train with opencl install the lastest Opencl torch lib:

![seq2seq](https://4.bp.blogspot.com/-aArS0l1pjHQ/Vjj71pKAaEI/AAAAAAAAAxE/Nvy1FSbD_Vs/s640/2TFstaticgraphic_alt-01.png)
_Source: http://googleresearch.blogspot.ca/2015/11/computer-respond-to-this-email.html_
```sh
luarocks install cltorch
luarocks install clnn
```

In this experiment, we train the seq2seq model with movie dialogs from the [Cornell Movie-Dialogs Corpus](http://www.mpi-sws.org/~cristian/Cornell_Movie-Dialogs_Corpus.html). The lines are shortened to the first sentence.
3. Download the [Cornell Movie-Dialogs Corpus](http://www.mpi-sws.org/~cristian/Cornell_Movie-Dialogs_Corpus.html) and extract all the files into data/cornell_movie_dialogs.

## Sample conversation

Basic Usage
===========
Here's a sample conversation after training for 20 epoch with 50000 examples, using the following command:

```sh
Expand All @@ -26,7 +41,7 @@ th train.lua --cuda --dataset 50000 --hiddenSize 1000

(Took 3 days to train on my GeForce GTX 780M.)

For OpenCL, use `--opencl` instead of `--cuda`. To train on CPU, don't provide any of those two.
For OpenCL, use `--opencl` instead of `--cuda`. To train on CPU, don't provide any of those two. Use the `--dataset NUMBER` option to control the size of the dataset. Training on the full dataset takes about 5h for a single epoch. The model will be saved to `data/model.t7` after each epoch if it has improved (error decreased).

> **me:** Hello?
> **bot:** Hi.
Expand Down Expand Up @@ -60,52 +75,6 @@ For OpenCL, use `--opencl` instead of `--cuda`. To train on CPU, don't provide a
>
> **me:** What are you?
> **bot:** I'm not sure.
>
> **me:** Do you plan on taking over the world?
> **bot:** No, i don't.
Phew! That was close. Good thing I didn't train it on the full dataset. Please experiment responsibly.

_(Disclaimer: nonsensical responses have been removed.)_

## Installing

1. [Install Torch](http://torch.ch/docs/getting-started.html).
2. Install the following additional Lua libs:

```sh
luarocks install nn
luarocks install rnn
luarocks install penlight
```

To train with CUDA install the latest CUDA drivers, toolkit and run:

```sh
luarocks install cutorch
luarocks install cunn
```

To train with opencl install the lastest Opencl torch lib:

```sh
luarocks install cltorch
luarocks install clnn
```

3. Download the [Cornell Movie-Dialogs Corpus](http://www.mpi-sws.org/~cristian/Cornell_Movie-Dialogs_Corpus.html) and extract all the files into data/cornell_movie_dialogs.

## Training

```sh
th train.lua [-h / options]
```

Use the `--dataset NUMBER` option to control the size of the dataset. Training on the full dataset takes about 5h for a single epoch.

The model will be saved to `data/model.t7` after each epoch if it has improved (error decreased).

## Testing
To load the model and have a conversation:

Expand All @@ -115,26 +84,6 @@ th -i eval.lua --cuda # Skip --cuda if you didn't train with it
th> say "Hello."
```

## License

MIT License

Copyright (c) 2016 Marc-Andre Cournoyer

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
Credits
===========
Credit for the vast majority of code here goes to [Marc-André Cournoyer](https://github.com/macournoyer). I've merely created a wrapper around all of the important functions to get people started.

0 comments on commit 86abfe8

Please sign in to comment.