Lean Restocking Alert Signal: Deployment

This workflow deploys the lean restocking system for bycicles in Washington DC stations, as trained in another workflow. Data is a subset of the bike dataset. The model will predict one of three classes: Add bikes; Remove bikes; No Action.
1. Data are read and prepared as in the training workflow, that is: Bike ratios are calculated as total # bikes/available spots by station; 10 past ratios in input vector
2. Model is read and applied
3. Prediction class (besides No Action) is accepted only if confidence exceeds a given threshold
4. Stations in need of bike reshuffling are shown in a web dashboard