Add possibility of limiting the prediction horizon

[christophme]

I wanted to predict short-term survival curves from a RSF on a large customer dataset and ran out of memory because the entire survival curves were predicted with their length equal to the training dataset.

It would be great if there would be the possibility to limit the prediction by passing an array with indices or something like that similar to what is implement in the lifelines package for predictions via passing the 'time' - parameter.

References and existing implementations
[(https://lifelines.readthedocs.io/en/latest/fitters/regression/CoxPHFitter.html#lifelines.fitters.coxph_fitter.SemiParametricPHFitter.predict_survival_function)]

[sebp]

Could you please clarify what you mean by "their length equal to the training dataset"?

If you call predict_survival_function it will return an array of StepFunction instances, which share the x attribute, but have different y values.

Do you have a huge number of unique training times such that x and y cannot be stored?

If not, you can evaluate each StepFunction at any time point(s) by calling the function. See for instance, this example from the user guide:

rsf_chf_funcs = rsf.predict_cumulative_hazard_function(
    va_x_test, return_array=False)
rsf_risk_scores = np.row_stack([chf(va_times) for chf in rsf_chf_funcs])

[christophme]

With "their length equal to the training dataset" I meant that the length of the returned arrays or StepFunctino are determined from the training times and this number was so large that x and y couldn't even be stored. Here, I would have appreciated the possibility in limiting the prediction to the next n number of periods or a list of predefined time periods.