Matrix shaped features issue

[aadrian92]

Hello,

I am trying to apply the gplearn functions to matrix shaped features (i.e. maybe you have the Close price of some assets at different points in time - the feature would be a matrix with the assets as columns and timestamp as index) Hence I am reshaping the matrix features in vectors and then for each custom operation that would need the matrix form I am reshaping back to matrix, applying the operation and then reshaping back to vector form, something like this: def func(x): mat = pd.DataFrame(x.reshape(col_len, -1).T) opmat = Feature(mat) opmat.ts_std() res = opmat.alpha.copy() return res.fillna(0).values.T.reshape(1,-1)[0]

You can see that I am using a global variable col_len which tells the number of columns (assets). The issue is that I would like to pickle the model and apply it on the same features with a different number of assets (so different col_len). But obviously the functions were created with a fixed parameter col_len. Is there anything you would suggest me to do to achieve this goal, please? I've tried to put col_len at the beginning of the input vector x or add delimiters to x were each row would end. Unfortunately it didn't work because the first value or delimiter was altered in the process (although I made sure they remain constant when applying the custom functions)

Thank you and I appreciate any help!

[trevorstephens]

The X's need to have the same features in both the training and testing sets, so doesn't sound like what you want to do is going to be possible sorry