Simple chatbot implementation with PyTorch.
- The implementation is straightforward with a Feed Forward Neural net with 2 hidden layers.
- Customization for your own use case is super easy. Just modify
intents.json
with possible patterns and responses and re-run the training (see below for more info).
The approach is inspired by this article and ported to PyTorch: https://chatbotsmagazine.com/contextual-chat-bots-with-tensorflow-4391749d0077.
Whatever you prefer (e.g. conda
or venv
)
mkdir myproject
$ cd myproject
$ python3 -m venv venv
Mac / Linux:
. venv/bin/activate
Windows:
venv\Scripts\activate
For Installation of PyTorch see official website.
You also need nltk
:
pip install nltk
If you get an error during the first run, you also need to install nltk.tokenize.punkt
:
Run this once in your terminal:
$ python
>>> import nltk
>>> nltk.download('punkt')
Run
python train.py
This will dump data.pth
file. And then run
python chat.py
Have a look at intents.json. You can customize it according to your own use case. Just define a new tag
, possible patterns
, and possible responses
for the chat bot. You have to re-run the training whenever this file is modified.
{
"intents": [
{
"tag": "greeting",
"patterns": [
"Hi",
"Hey",
"How are you",
"Is anyone there?",
"Hello",
"Good day"
],
"responses": [
"Hey :-)",
"Hello, thanks for visiting",
"Hi there, what can I do for you?",
"Hi there, how can I help?"
]
},
...
]
}