Skip to content

Troubleshooting

ModuleNotFoundError: No module named 'gen_surv'

The package is not installed in the interpreter you are running:

pip install gen-surv

From a source checkout, install it rather than relying on the working directory:

pip install -e .      # or: poetry install

Check which interpreter is actually in use — a mismatch between the shell's python and your editor's is the usual cause:

import sys; print(sys.executable)

ImportError: cannot import name 'to_sksurv' from 'gen_surv'

scikit-survival is the one optional dependency, and those two helpers are only exported when it is present:

pip install scikit-survival

If the build fails, conda-forge has prebuilt wheels: conda install -c conda-forge scikit-survival. See Installation.

ChoiceError: Argument 'model' must be one of ...

The model name is not one of the eleven. The message lists them all; note aft_ln rather than aft_log_normal, and piecewise_exponential rather than piecewise.

LengthError: Argument 'beta' must be a sequence of length 3

Several models require an exact number of coefficients:

Model beta length Also
cphm a scalar, not a sequence —
cmm, thmm exactly 3 rate is 6 for cmm, 3 for thmm
tdcm exactly 2 3 is deprecated and warns
aft_* any length — it sets the covariate count —

The error names the model it was validating, which tells you which rule applies:

LengthError: Argument 'beta' must be a sequence of length 3; got length 2.
Adjust the number of elements. (while validating inputs for model 'thmm')

DeprecationWarning: gen_tdcm uses two coefficients

You passed three; the third has never had any effect. Drop it — see TDCM.

ParameterError: ... must include 'mean' and 'std'

covariate_params has to be complete for its distribution; there is no per-key defaulting. Pass all of the required keys, or pass None to take the defaults. See Covariates.

The frame has more rows than n

Expected for cmm and thmm: subjects contribute two or three rows each. Use df["id"].nunique(), not len(df). See Output schemas.

KeyError: 'time' on a tdcm or thmm frame

Column names differ by model. tdcm has start/stop rather than time, and thmm has time/state with no status at all. Most helpers accept time_col and status_col:

from gen_surv import describe_survival, generate

tdcm_df = generate(model="tdcm", n=100, dist="weibull", corr=0.5,
                   dist_par=[1.0, 2.0, 1.0, 2.0], model_cens="uniform",
                   cens_par=5.0, beta=[0.5, 0.3], lam=1.0, seed=1)

describe_survival(tdcm_df, time_col="stop")

Everything is censored, or nothing is

cens_par is an upper bound for uniform censoring and a mean for exponential censoring, so larger values censor less. If the event rate is 0.0 or 1.0, that dial is at the wrong end — sweep it, as in Censoring.

An estimator does not recover the parameter I set

Work through these in order:

  1. Sample size. At n=2000 a Cox estimate can miss the truth by two standard errors — that is ordinary variation. Repeat over seeds and look at the mean, or raise n.
  2. The parameter means what you think. beta in aft_weibull is a log hazard ratio; its effect on log time is \(-\beta/\texttt{shape}\). See AFT models.
  3. The estimator matches the model. A Fine-Gray fit will not return the cause-specific betas used to generate competing-risks data — by design.
  4. The coefficients were not random. With betas=None, several models draw the coefficients for you, so there is no fixed truth to recover. Always pass them explicitly when validating.
  5. The layout suits the estimator. A naive Cox fit on tdcm output is biased, and flips the sign, because the risk interval is not split at the crossover. Split it using the crossover time from simulate() — see TDCM.

Results changed after upgrading

A bug fix in a sampler changes the numbers a given seed produces; 1.3.0 and 2.0.0 both did this deliberately. 3.0.0 did it for cmm and thmm specifically, by rebuilding both on the multistate engine. Pin the version alongside the seed for anything that must reproduce. See Reproducibility.

Plots do not appear

In a script or on CI there is no display. Save instead of showing:

import matplotlib
matplotlib.use("Agg")

from gen_surv import generate, plot_survival_curve

df = generate(model="cphm", n=200, beta=0.5, covariate_range=2.0,
              model_cens="uniform", cens_par=1.0, seed=1)
fig, ax = plot_survival_curve(df)
fig.savefig("plot.png", dpi=200, bbox_inches="tight")

Something else

Search or open an issue on the tracker. A minimal snippet including the gen_surv.__version__ and the seed is enough to reproduce almost anything here.