/intent-deeppavlov

Simple intent recognition using DeepPavlov.ai

Primary LanguageDockerfile

Simple intent detection service based on DeepPavlov.ai.

This should be used only if you have large amount of training data

Training is happening on image build event. Configuration should be similar to https://doc.livehelperchat.com/docs/bot/deeppavlov-faq/ except response location should be 0:0 instead of 0:0:0

  • You should edit Dockerfiles/deep/train/train.json with your data before building an image.
  • See Dockerfiles/deep/Dockerfile for getting dstc2_fastText_model.bin file
docker-compose -f docker-compose.yml build
docker-compose -f docker-compose.yml up

Run as service once it's build

docker-compose -f docker-compose.yml up -d

Testing

curl -X POST "http://localhost:5000/model" -H "accept: application/json" -H "Content-Type: application/json" -d "{\"x\":[\"I want to book place for two\"]}"

Response if you did not change train data

[["BookRestaurant",[0.2782818078994751,0.22770161926746368,0.30094414949417114,0.6002139449119568,0.2533802390098572,0.25423356890678406,0.21291431784629822],3]]