You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
1. Regarding predit_surv in line 155 surv = 1. - cif.sum(0)
why it needs to take the sum for cif? The cumulative sum has already been calculated in cif, and in my view, it should be: surv = 1 - cif
2. Regarding predict_pmf in line 202 pmf = pad_col(preds.view(preds.size(0), -1)).softmax(1)[:, :-1]
what are the intuitions for padding another column for the model output before softmax?
In the original paper "Continuous and Discrete-Time Survival Prediction with Neural Networks", page 7, it says Alternatively, one could let φm+1(x) vary freely, something that is quite common in machine learning, but we chose to follow the typical conventions in statistics.
I am confused about this trick. If I took the φm+1(x) very large, then the estimated probability (after applying softmax) for each time interval would be very small, will this affect my predictions?
Any comments would be appreciated! Thanks in advance!
The text was updated successfully, but these errors were encountered:
Thanks for creating this wonderful package!
I have a few questions regrading the functions for deephit model:
https://github.com/havakv/pycox/blob/master/pycox/models/deephit.py
1. Regarding
predit_surv
in line 155surv = 1. - cif.sum(0)
why it needs to take the sum for
cif
? The cumulative sum has already been calculated in cif, and in my view, it should be:surv = 1 - cif
2. Regarding
predict_pmf
in line 202pmf = pad_col(preds.view(preds.size(0), -1)).softmax(1)[:, :-1]
what are the intuitions for padding another column for the model output before softmax?
In the original paper "Continuous and Discrete-Time Survival Prediction with Neural Networks", page 7, it says
Alternatively, one could let φm+1(x) vary freely, something that is quite common in machine learning, but we chose to follow the typical conventions in statistics.
I am confused about this trick. If I took the φm+1(x) very large, then the estimated probability (after applying softmax) for each time interval would be very small, will this affect my predictions?
Any comments would be appreciated! Thanks in advance!
The text was updated successfully, but these errors were encountered: