- Coupon day data improves the MSE
- MSE for lower/upper uncertainity interval improves if we reduce the interval width (to 0.5 vs 0.8 default)
- Need a lot more data for using 'logistic' growth (with current data, predictions are much higher)
- The predictions MSE improves in general with more data
- The lower/upper MSE are not deterministic (they keep changing slightly with every run, with no change in anything else)
- lower/upper CAP doesnt affect any predictions yet (since we are using high limits)
TODO
- Channel/stream API - End point which consumes the time series points -
{"ds": "<time stamp>", "value": "X"}
- Channel/stream API requires auth token or secret, these token/secrets will be stored in db
- A page to login:
- list channels
- see channel data + predictions
- create/delete a channel
- create/delete a token/secret
- Data for each channel stored in redis
- cron script to send predictions to emails ?
- query predictions from slack by channel name ?
- python and flask