[Feature]: can I design a custom operator for sequence data?

[callmedxx]

Feature Request

I want to use SymbolicRegression.jl to analyze stock market data. Specifically, I need to design a rolling operator to process time series data, similar to rolling_sum in pandas. The input of this operator consists of time series data, the window size, and the sliding stride required for rolling. How can I design such an operator for SymbolicRegression.jl?

[MilesCranmer]

By input to the operator, do you mean input to the expression?

You can learn expressions over time series without extra effort by simply repeating/reshaping your time series data to have shape:

(window_size * num_features, num_examples)

if you have three variables in your time series say x y and z, and your window size is 2, then you would have features x_1, y_1, z_1, x_2, y_2, z_2 - where x_2 is simply the previous value of x.

For this sort of reshaping I recommend https://github.com/mcabbott/TensorCast.jl