Conversation
hfrick
left a comment
hfrick
left a comment
There was a problem hiding this comment.
Some more small edits as suggestions 😄
|
|
||
| - Observed time: time recorded in the data | ||
| - Event time: observed times for actual events | ||
| - Evaluation time: the time, specified by the analyst, that the model is evaluated. |
There was a problem hiding this comment.
should this be "that the model is evaluated at"? or maybe "evaluated for"? 🤔
|
Todo:
|
| - **Category 1 - Events**: Evaluation time is greater than or equal to the event time ("it has already happened"). | ||
| - **Category 2 - Non-events**: Evaluation time is less than the observed time, censored or not ("nothing has happened yet"). |
There was a problem hiding this comment.
I've removed the "definitive" qualifier here since we then go on to add a probability for that and that feels somewhat "off" to "definitive".
There was a problem hiding this comment.
Another important change: I've flipped the language and now describe it in terms of evaluation time relative to observed time since evaluation time is the thing we are varying and the observed time is fixed.
| geom_vline(aes(xintercept = eval_time, col = I("red"), linetype = I("dashed"), linewidth = I(0.8))) + | ||
| geom_point(aes(eval_time, obs_id, shape = eval_status, col = I("red"), size = I(5))) + |
There was a problem hiding this comment.
I have not figured out in justifiable time how to get the red into the legend. Accept that small mismatch or go back to all black?
not using `.time_as_binary_event()` since that requires a single value for eval_time
| binary_encoding <- | ||
| dyn_val_pred %>% | ||
| mutate( | ||
| obs_class = time_as_binary_event(event_time, .eval_time), |
There was a problem hiding this comment.
.time_as_binary_event() from parsnip wants eval_time as a single numeric value, to use that it would have to be something like this:
binary_encoding <-
dyn_val_pred %>%
rowwise() %>%
mutate(
obs_class = .time_as_binary_event(event_time, .eval_time),
) %>%
ungroup() %>%
mutate(
pred_class = if_else(.pred_survival >= 1 / 2, "non-event", "event"),
pred_class = factor(pred_class, levels = c("event", "non-event")),
)
So if we want to remove the function definition here and use the parsnip version, we might want to consider tweaking that one first. The above code block feels pretty heavy on the friction.
2 participants
I'll leave this as draft until the other articles are also drafted.