weibull_aft.plot_survival_function: object has no attribute

[timscorbett]

Is this a known issue? Is there a workaround?

Traceback (most recent call last): File "lifelines_example.py", line 31, in weibull_aft.plot_survival_function(ax=axes[0][0]) AttributeError: 'WeibullAFTFitter' object has no attribute 'plot_survival_function'

[CamDavidsonPilon]

What is the survival function of an AFT model? The AFT model is conditional (i.e requires covariates). Maybe you want predict_survival_function?

[timscorbett]

I see an example here on page 11-12 https://buildmedia.readthedocs.org/media/pdf/lifelines/latest/lifelines.pdf

[CamDavidsonPilon]

That's the WeibullFitter, not WeibullAFTFitter

[timscorbett]

Ah! thank you. How can I get the baseline curve alone? At this time, my only covariate is a continuous one. I will use the partial feature after augmenting the dataset.

On Thu, Sep 28, 2023 at 5:26 PM Cameron Davidson-Pilon < @.***> wrote:

That's the WeibullFitter, not WeibullAFTFitter

— Reply to this email directly, view it on GitHub https://github.com/CamDavidsonPilon/lifelines/issues/1562#issuecomment-1740154970, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANQJMZPRCFFESSINWQOOPODX4YIT7ANCNFSM6AAAAAA5LPF4UI . You are receiving this because you authored the thread.Message ID: @.***>

[CamDavidsonPilon]

How can I get the baseline curve alone

There's not really a baseline curve for AFT models (terminology isn't used), but I think you just want to set the covariate to 0 in predict_survival_function.

[timscorbett]

Never mind, I added a relevant constant column T.

• How do I get the CDF relative to this conditional variable from plot_partial_effects_on_outcome? 1 - weibull_aft.predict_survival_function('T') ?

• Where can I give the T range as given in plot_partial_effects_on_outcome?

• As well how do I say, I need these for times t=1..10 ?

• I see predict_survival_function() gives me as many curves as there are rows in the df. Can there be one prediction per T and one single prediction for the baseline?

[CamDavidsonPilon]

The first three questions can be answered by checking out the docs: https://lifelines.readthedocs.io/en/latest/fitters/regression/WeibullAFTFitter.html?highlight=plot_partial_effects_on_outcome#lifelines.fitters.weibull_aft_fitter.WeibullAFTFitter.plot_partial_effects_on_outcome

I don't quite understand your 4th question, however

[timscorbett]

That helps. Thanks Cam!

I guess I can get the cumulative distribution function (CDF) as 1-survival from the plot collection of partials.

Is there a confidence interval band fill between for the survival plot? If not, how can I get it?

[CamDavidsonPilon]

Is there a confidence interval band fill between for the survival plot? If not, how can I get it?

Unfortunately, not

[timscorbett]

Under what conditions is the cumulative_hazard < hazard? My plots have t = 0 to 6. The Y for hazard rate goes up to 750 at t=6. THe Y for cumulative hazard goes only up to 450 at t=6 The survival plot looks as expected..

On Sat, Sep 30, 2023 at 3:15 PM Cameron Davidson-Pilon < @.***> wrote:

Is there a confidence interval band fill between for the survival plot? If not, how can I get it?

Unfortunately, not

— Reply to this email directly, view it on GitHub https://github.com/CamDavidsonPilon/lifelines/issues/1562#issuecomment-1741870723, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANQJMZNT4D363SP4OXZKTPDX5CKWVANCNFSM6AAAAAA5LPF4UI . You are receiving this because you authored the thread.Message ID: @.***>

[CamDavidsonPilon]

cumulative hazard (t) = int_0^t hazard(s) ds

so it's possible for cumulative_hazard < hazard. Think about how a short spike in a function might affect its integral.

[timscorbett]

Thanks Cam. If my input data itself is in log10 and the shape is:

                         coef      exp(coef)

rho_ Intercept 2 9

Is it safe to say, my antilog(rho_) for interpretation is 10^(2) and not 10^(9)?

On Tue, Oct 3, 2023 at 2:19 PM Cameron Davidson-Pilon < @.***> wrote:

cumulative hazard (t) = int_0^t hazard(s) ds

so it's possible for cumulative_hazard < hazard. Think about how a short spike in a function might affect its integral.

— Reply to this email directly, view it on GitHub https://github.com/CamDavidsonPilon/lifelines/issues/1562#issuecomment-1745742074, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANQJMZOHNEKIIZ4BQKY3MHDX5R6NDAVCNFSM6AAAAAA5LPF4UKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONBVG42DEMBXGQ . You are receiving this because you authored the thread.Message ID: @.***>

[CamDavidsonPilon]

Sure, yea, but you should expect very small variation in your log10 variable as a consequence.

[timscorbett]

Is the plot_partial_effects_on_outcome same as a regular weibull(1.75,2.24) plot conditional on T? Here shape (rho_) = 2.24 and scale (lambda_) = 1.75; based on the results below My intent is to get upper and lower confidence interval curves for this partial effect based on the interval you provide for rho_.

(By the way, there is some literature on confidence intervals for Cox PH survival with covariates. Not sure if those are conditional on a covariate. https://www.jstor.org/stable/2530904)

[timscorbett]

For WeibullAFT, why does not predict_survival_function have partial outcomes?

On Thu, Oct 5, 2023 at 3:51 PM Cameron Davidson-Pilon < @.***> wrote:

Sure, yea, but you should expect very small variation in your log10 variable as a consequence.

— Reply to this email directly, view it on GitHub https://github.com/CamDavidsonPilon/lifelines/issues/1562#issuecomment-1749761816, or unsubscribe https://github.com/notifications/unsubscribe-auth/ANQJMZOGANFWD5USMNRK67DX542WXAVCNFSM6AAAAAA5LPF4UKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONBZG43DCOBRGY . You are receiving this because you authored the thread.Message ID: @.***>