Hi, would you like to add some examples for multiple-symbol-runs?

[narcissusdu]

We see ts_corr might be unable to work out the correlation for multiple-symbol scenarios.
Is that supported?

Thanks,

[narcissusdu]

We see a similar issue closed. Do you have any furthur updates for that?
#2

[UePG-21]

This is originally designed for single symbol, but you may add other symbol data as raw factors.

[narcissusdu]

Thanks for the reply!
We have tried that. It works good with simple calculations such as mul/div/abs.
The difficulty we found was that gplearn only supports functions for ndarrays. It erases the 2 index we added (symbol/date). Unless we convert the ndarry back to 2d array inside the function with some tricks, it won't get a meaningful corr.
Do you have any suggestions?

[uepgwang]

ts_corr should be able to take two 1d array and calculate correlation. I don't know why it doesn't work.

[narcissusdu]

With ts_corr(x1, x2, d) we are tring to calculate the previouse d-days' ts_corr of 2 indicator-data within the same symbol.
In single symbol scenario it works perfectly. While in multiple-symbol scenario, ts_corr needs to use a subsequence of the 1d array which have the same symbol_index, to calculate the previous d-day corr. Or the corr could be calculated across different symbols and different days.

Given an 2-index dataframe with 2 symbols

day_index	symbol_index	open	close
1	a	open_1a	close_1a
1	b	open_1b	close_1b
2	a	open_2a	close_2a
2	b	open_2b	close_2b
3	a	open_3a	close_3a
...

If a generation was created as ts_corr(open array, close array, 3) to such 2-index data, the calculation to the elements would have inconsistent meaning.