Cameron Davidson-Pilon

Hi, I'm finally investigating this more carefully. I've written these two tests below, but am unable to make them fail (all assertions are true). ```python def test_baseline_prediction_with_extreme_means(self, rossi): cph =...

@bofeng2018 email attachments won't show in issues, can you repost your nb in the issue thread?

(Sorry about the spam - I keep alternating between if I am convinced I've found the problem or not - still investigating)

Okay, here's an (updating) summary of what I know. This may change: - the baseline hazard (and cumulative hazard) can be computed at the mean values, or at all zeros....

Some more thoughts: - this `exp(-params * means)` factor is _introduced_ in the predict functions. So when you call `predict_survival_function(x)`, a `exp(-params * means)` is introduced. At a high level:...

Here's my edits of your notebook, I made some changes to reduce the complexity and focus on a simpler example. [Demonstration 2.ipynb.zip](https://github.com/CamDavidsonPilon/lifelines/files/13855860/Demonstration.2.ipynb.zip) ----- > To be more specific, covariate means...

> Overall, very inspiring discussions! Thanks for all the thought sharing! 💯 This is forcing me to re-evaluate these very important details. We'll make some changes to make this easier...

@Unessam thanks for volunteering - I think an example would be sufficient that demonstrates the interaction between `times` and `conditional_after`

Can you post a stacktrace and version of lifelines you are using?

Subtracting the mean, or any value really, doesn't change the final inference. We subtract the mean internally for stability reasons during fitting.