Profanity Filters Bot

• A demo can be found on Youtube
• Link to devpost here
• [NSFW] Release all you can say via Swagger under /predict
• biLSTM trained on iMDB for sentiment analysis
• torchtext was built from master, which can be found here

instructions

make help for further instructions

• prepare pretrained embeddings do make init
• To package the model to bento do make build
• to run the docker container do make run
• endpoint can be accessed through POST request via localhost:5000

curl -X POST 0.0.0.0:5000/predict -H "accept: */*" -H "Content-Type: application/json" \
-d "{\"text\":\"I hate you this is the worst experience I have ever seen\"}"
# > 0.2898484170436859

building torchtext from source

• do git clone --recurse-submodules https://github.com/MLH-Fellowship/1.0.5-DiscoBot && cd torchtext
• git submodules update --init --recursive && python setup.py clean install to build torchtext from source (currently at 0.8.0a0+8dc2125)
  • to prep the pretrained embedding run make prep, otherwise if you want to train the model parses ARGS=--train, like so:

# This will train the model
make ARGS=--train prep