sambucini

Hey all! let's say you have a lot data-sciency models in production (that for example forecast stuff) where each model relies on a bunch of hard-coded config parameters that you've found through some e.g. hyperparameter tuning etc. which you dont wanna rerun every time the production model runs. I end up hard-coding these values either directly in the production model code or put them in an enum/json somewhere. However, I wish I had some "config parameter management tool" with a pretty GUI and API that allowed me to store the parameters in a central database which a few features:

- allow for programmatically writing new parameters

- allow the production models to programmatically query the parameters they're supposed to use

- do versioning on the parameter sets, as it's useful to keep the history of parameters for backtesting/reporting

- have some logic of structuring the parameter sets thrugh grouping/tagging and/or a hierarchical model so all sets are not just one big mess

- finally allow of course for different parameter set structures, as different models have different parameters

- ideally a GUI

I didn't manage to find anything like it online, are you aware of anything? How are you dealing with that topic if it's an issue for you as well?

thanks a lot in advance!

Dear all,

time series DBs are a hot topic these days but given the focus on IoT -- or more generally measurement data -- the underlying data model typically assumes that for one time series there is only one data point per period of time, so it's really just one dimensional. However, if you work for example with forecast data (say for a stock price) you might wanna store every version of a forecast and not overwrite the previous forecast. What are in your experience the best time series databases that (natively) support two or more dimensions and also allowing queries on these other dimensions like "get forecast for delivery-time from x to y where forecast_time = z"?

Thanks! sambucini