This is a project of Samurai Technology. Contact us at hi@samurai.technology.

Smart Popup provides an advice whether to show or not a pop-up. The advice is based on ML models trained on website's users activity data. Smart Popup currently supports 2 models:

• Initial model: model based on user initial data, which can be recorded when a user enters a website
• Activity model: model based on initial model data and user activity events

Production models of Smart Popup are not open sourced as they are created based on a client's data. Dummy models are provided for test and tutorial purposes.

To run Smart Popup with custom models:

1. Serialize your models with directory pickle.
2. Configure path to serialized models in App/config.ini file (MODELS)
3. Register deserialized models in ModelsRegistry before starting the server.
4. Optional: populate database with custom data (constants), for instance: discrete values from model's input variable space (e.g. device_categories = ["desktop", "mobile", "tablet"]). See tests/dbPopulator/DBPopulator.py for an example. This data can be used while utilizing the models during the request.
5. Provide predictors for initial and activity requests. Predictor is a function from data populated in point 3., request body and ModelsRegistry into an advice for the request. Register predictors in PredictorsRegistry and use them in DecisionService. Predictor.predict(...) should be a pure function. See predictors package for an example.

1. Create docker network: ./bin/docker-create-network.sh
2. Create backend image: ./bin/docker-build-backend.sh
3. Run database: ./bin/docker-run-database.sh
4. Run backend: ./bin/docker-run-backend.sh

Server will be available at localhost:8888.

Smart Popup can be used to update a flag deciding whether to show a popup to a current website's user. Workflow schema:

While installing new dependencies, remember to update app/requirements.txt file.