Support Time Series Forecasting

[joelchen]

Would like to forecast future record(s) of CSV with Tangram based on existing data or a subset of existing data that is close to the current time (https://www.tensorflow.org/tutorials/structured_data/time_series).

[nitsky]

Hi @joelchen thank you for opening this issue, this is a very good idea and definitely something we want to support. Can you provide some specifics on what you would like the interface to look like?

[joelchen]

Thank you @nitsky for getting back. Simple interface will be preferred. An example can be referenced from https://github.com/mindsdb/mindsdb-examples/blob/master/others/us_consumption/mindsdb_acc.py (https://docs.mindsdb.com/tutorials/Timeseries).

[samgaw]

I was asking Isabella about this. My two cents:

• At a minimum, ARIMA style predictions. Maybe krfricke/arima would be a good starting point?
• The ideal would be an ARIMA model that accepts Fourier terms to allow for seasonality. Pivotal wrote an example guide a while back on this using MADlib.
• The massive, unrealistic version would be something akin to Prophet that takes a guess at seasonal breaks in the data and takes an additive approach.