0

I've built a model using PySurvival's CoxPH model and then ran the predict_survival function to check its output. Here's a sample of the output it provides when printing the result of the predict_survival function after calling it with the following parameters:

coxph = CoxPHModel()
coxph.fit(train_X, train_T, train_E, lr=0.5, l2_reg=1e-2, init_method='zeros')

 [9.96254322e-001 9.86763222e-001 9.47647159e-001 6.05518522e-001
  3.54227963e-001 7.71126785e-002 2.95367274e-022 0.00000000e+000
  0.00000000e+000 0.00000000e+000 0.00000000e+000 0.00000000e+000
  0.00000000e+000 0.00000000e+000 0.00000000e+000 0.00000000e+000
  0.00000000e+000 0.00000000e+000 0.00000000e+000]

As far as my understanding goes, each of the numbers in the above array are a percentage stating the odds of my event of interest occurring on a date but I don't understand what that date is in relation to the array or how I can extrapolate this info to run it on input different than my test input. Also every array output by the survival functions only contains 15 elements and the odds almost always hit 0 around the 10th element. I just don't get how I can best interpret this data so that I can use the results elsewhere.

Zach Goodman
  • 1
  • 1
  • If the design of `CoxPH.predict_survival` is anything like R's `survival::predict.coxph` there are parameters that would need to be known for any interpretation to be offered. So including the calling code in the body of your question by using SO's [edit] would be needed. – IRTFM Jan 06 '21 at 17:10
  • Just edited my initial post to include the parameters I'm using. – Zach Goodman Jan 06 '21 at 18:59

1 Answers1

0

Your training data has survival time of max 15 days. Hence, it learns the probability of survival for the 15 days and outputs accordingly.

Based on the condition of the data, it seems to be having 0 chances of survival post the 10th day.

das1
  • 1
  • 1