This project is intended to create an automatic reply system or ChatBot. It has been trained for twitter and cornell-movie dataset and currently supports both of them. This project is the first version of more advance bot system project I am working upon.
The working of the automatic repy systems in online messengers and services fascinated me. In order to have more knowledge about their working, the basic need was to create a project to understand the basic needs of such a model and how to train it. The twitter and movie dataset is taken from their respective sources and are available under data. The coding style and motivation is found from seq-chatbot. Tensorlayer is used as a high level wrapper over the tensorflow backend.
The project is created from Sequence to Sequence model and consists of two RNNs - an encoder and a decoder. The encoder reads the input sequence, word by word and emits a context, which would ideally capture the essence of the input sequence. Based on this context, the decoder generates the output sequence, one word at a time while looking at the context and the previous word during each timestep.
RNN are Recurrent Neural Networks in which nodes form a graph,hence they can be used to identify sequences in a text, as contrast with Feed Forward NN where there is no cycle.
To run the application choose the required model. Modelv1t.npz is 1 epoch train model on twitter dataset. Modelv2t.npz is 15 epoch train model on twitter dataset and Modelcopper.npz is 15 epoch train model on cornel-movie dataset. Change the name in script accordingly.
- To install the necessary requirements, follow the procedure
pip install -r requirements.txt- Launch 'run.py' script to get query results provided in seeds.
python run.py- To train the model on the required dataset launch 'main.py' script.
python main.py| INPUT QUERY | OUTPUT QUERY |
|---|---|
| How is the weather? | i think it was a good one thing |
| How are you? | i am so excited for the next time |
| I had an exam today. | thank you |
| Do you have any political views? | i am so excited for the debate |
| Tell me something about yourself | i think its a lot of people |
| happy birthday have a nice day | thank you |
The 'main.py' script launches as:
The time taken to run it is quite large.
The 'run.py' script launches as:
- Python- 3.6
- Tensorflow- 2.0.0b0
- Tensorlayer- 2.0.2
The code working is provided in comments in the respective scripts.