Add some classic time-series

[microprediction]

Add some classic time-series in loops. But do it in a way that it takes the data a LONG time to exhaust completely. What I have in mind in the following:

1. Choose a long univariate series from Libra, UCI, UC whatever
2. Or take two, and hit with a 2 x 2 matrix.
3. Hold out the second half the data set
4. Loop over the first half
5. On the next loop, include one data point (or a few) from the second half of the series
6. Etc

Maybe arrange it so that the extra point (or a few) enter at a specific time of day, so if anyone really cares, there could be an assessment completely out of sample.

Or maybe there are two streams, and one stream only ever had new points added.

Since these are usually public, we probably shouldn't obsess too much, since people can cheat if they really have their hearts set on it.

[microprediction]

Libra paper

[microprediction]

As a bonus this could be a nice experiment in information. After all, one could reconstruct predictions by inverting the 2x2 matrix. I wonder how much that hinders the algos (or helps??)

[microprediction]

I suspect PandasLoop might be rather helpful, though that doesn't have the "add one more in" capability. Maybe create CaterpillarLoop derived from it, or just a separate class.

I think cset requires MUID strength 13 btw

[microprediction]

https://github.com/DescartesResearch/ForecastBenchmark/blob/main/R/sysdata.rda

[microprediction]

Data is also here: https://zenodo.org/record/4399959#.YWWfvBDMK6A

[microprediction]

Too easy! pd.read_csv('https://zenodo.org/record/4399959/files/economics_11.csv?download=1') probably works

[DrAndreBauer]

Hi, thanks for reaching out. I somehow didn't get a notification. If there is still interest, we can collaborate.

[microprediction]

Hi Andre,

No worries I long since stopped noticing github alerts as they are buried under all my system failures :)

By way of intro, it seems we have a shared interest in benchmarking and you have done some fine work. My interest is twofold. First is finding good stuff for my hedge fund. Second is trying to pursue the long-term vision in my book. There are some notes above about possible obvious cross-use of our work, with the most obvious being adding some of your collected time-series to the live streams on the microprediction platform. Conversely it is pretty easy to pull data from microprediction if you are looking for more data (see here).

Peter

[DrAndreBauer]

Hi Peter,

thanks for reaching out here via mail.

Is this working with the hedge fund? I was always tempted to try something, but I’m too afraid to make mistakes 😃

Btw. I like the idea of the platform.

I would like to contribute as are both interested in benchmarking and also FAIR data. So far as I understand, the data there is kind of live data. Or do you also have “non-live” data?

Best,

André

Von: Peter Cotton @.> Gesendet: Donnerstag, 18. August 2022 16:22 An: microprediction/timemachines @.> Cc: DrAndreBauer @.>; Comment @.> Betreff: Re: [microprediction/timemachines] Add some classic time-series (#45)

Hi Andre,

No worries I long since stopped noticing github alerts as they are buried under all my system failures :)

By way of intro, it seems we have a shared interest in benchmarking and you have done some fine work. My interest is twofold. First is finding good stuff for my hedge fund. Second is trying to pursue the long-term vision in my book https://mitpress.mit.edu/9780262047326/microprediction/ . There are some notes above about possible obvious cross-use of our work, with the most obvious being adding some of your collected time-series to the live streams https://www.microprediction.org/browse_streams.html on the microprediction platform.

Peter

— Reply to this email directly, view it on GitHub https://github.com/microprediction/timemachines/issues/45#issuecomment-1219556093 , or unsubscribe https://github.com/notifications/unsubscribe-auth/AG2INBX2OBQCY274NX7AFW3VZZBKHANCNFSM5F2WB4JQ . You are receiving this because you commented.Message ID: @.***>