This is the webapp that shows the results of brand predictions in a webapp. Unfortunately the pickled model (all_models_beauty.pkl) could not be uploaded because it was too large.
Files for getting products:
- allProducts_leverbaar2.pig is a pig script for getting all Products that are deliverable from an HBase storage
- pyUDF4.py is for cleaning up descriptions with very raw text
Files for creating models:
- create_model_beauty_and_health.ipynb ipython notebook to build a simple model (support vector machine) based on the product title
- beauty_leverbaar_20160729.txt simple data file to be used for the model fitting
Files for webapp:
- index.html is the frontend of the webapp for predictions
- folder /Static contains all css / js (Angular) for the webapp
- brand_predict.py is launching the Flask service (locally) that also makes predictions per brand
- brand_translator2.pkl is a dictionary that translates a brand_id to a brand_name (to make predictions more readable)
Extra files:
- 20161020 - Mooi Gezond.ipynb Get some insights in how the model performs
- 20161025 Separate model for every category.ipynb Build a model for all categories. This works for some but not all categories