👋 Hi folks, I'm the reviewer for your JOSS submission. I'll be making issues in this repo for my feedback.

1. Currently, github actions will only run automated tests (CI) on pull requests. I would suggest making this on pull_request and push. Contributors will push to the main branch, and you want CI to run for these commits, too.
2. There is installation instructions to install from pypi, but what about locally? I git pulled the repo, ran pip3 install -e ., and the py.test to test. That can be documented in a section on the README or the docs.
3. tests in test_msm_examples.py and test_sim.py should assert some output. There should be a reason you are running a test - assert that.
4. This can be fixed (found running py.test on my machine, py3.9.6)

src/pymsm/datasets/__init__.py:319
  /Users/camerondavidson-pilon/code/pymsm/src/pymsm/datasets/__init__.py:319: DeprecationWarning: invalid escape sequence \R
    1: "Discharged\Recovered",

5. If possible, try to fix the lifelines warnings that show up. Not always possible, I know.
6. stepfunc seems like an important function, but has lots of potential edge cases. This would be a good function to test as well.

From the checklist:

Are there clear guidelines for third parties wishing to 1) Contribute to the software 2) Report issues or problems with the software 3) Seek support

Typically project authors will include a CONTRIBUTING.md with this information.

• typo in Costume -> Custom
• How does it compare to software outside of Python, too?

Dear all,

Thanks for making this package available.

While running a Montecarlo simulation with model.run_monte_carlo_simultion, I found that the output trajectories are either unsorted or the times represents the time between events.

As an example:

    simulated_paths = model.run_monte_carlo_simulation(
        sample_covariates=covariates,
        origin_state=0,
        current_time=0,
        n_random_samples=100,
        max_transitions=3,
        n_jobs=5,
        print_paths=True,
    )

the outcome is

States: [3, 4, 6]
Transition times: [28.000029122914018, 912.999970877086, 1451.0]
States: [3, 6]
Transition times: [32.000013949386066, 2359.999986050614]
States: [3, 4, 1]
Transition times: [26.000044015249372, 425.9999559847506, 366.0]

in the last of the path reported above, last time_at_each_state is lower than the preceding ($366<425$).
Does this imply:

• the output of the simulation in path.states are unsorted and need to be sorted according to path.time_at_each_state;
• the times in path.time_at_each_state represent the time between states, hence we need to compute the cumulative time ourselves.

Finally I'm doubting about the structure of the input date for Pathobject: in states I read:

States visited (encoded as positive integers, 0 is saved for censoring), in the order visited. Defaults to None.

hence I suppose that time_at_each_state should also be ordered but maybe it should be a time difference between states?

Some of these are easy tasks, some are just suggestions to provide better docs.

1. The github action badges on the README.md point to the .svg image, but should point to the action (ex: https://github.com/hrossman/pymsm/actions/workflows/tests.yml instead of https://github.com/hrossman/pymsm/actions/workflows/tests.yml/badge.svg)
2. Running the code here did give an error at the end (a lifelines convergence error - can these be caught and report back to the user something more useful?)
3. In this intro, can you describe what a survival MSM is and why one might want to use it? How does it fit into the survival analysis toolkit?
4. Maybe a quick explanation or interpretation of the plot on this page. How should I read it?
5. In examining a model, what should I, the naive reader, be looking for? (BTW I like the color scheme you picked for the stackplot)
6. The full examples are nice. I think you could add more context about what the dataset is describing, how you are modelling it / choices you make, and conclusions you (the experts) draw from it).
7. Some broken latex on this page

Overall, I like your docs! I would like more emphasis on educating newbies. Y'all are experts, teach me!

I have prepared a pandas dataframe according to the documentation https://hrossman.github.io/pymsm/usage/Preparing_a_dataset/#preparing-a-dataset-for-multistate-modeling-with-pymsm but I receive this error message when I try to fit the model:

Init MultistateModel

from pymsm.multi_state_competing_risks_model import MultiStateModel
multi_state_model = MultiStateModel(
dataset=dfML,
covariate_names=covariate_cols,
terminal_states=terminal_states)

AttributeError Traceback (most recent call last)
Cell In[64], line 3
1 # Init MultistateModel
2 from pymsm.multi_state_competing_risks_model import MultiStateModel
----> 3 multi_state_model = MultiStateModel(
4 dataset=dfML,
5 covariate_names=covariate_cols,
6 terminal_states=terminal_states)

File ~\AppData\Roaming\Python\Python311\site-packages\pymsm\multi_state_competing_risks_model.py:127, in MultiStateModel.init(self, dataset, terminal_states, update_covariates_fn, covariate_names, event_specific_fitter, competing_risk_data_format, state_labels, trim_transitions_threshold)
125 self._trim_transitions()
126 else:
--> 127 self._assert_valid_input()

File ~\AppData\Roaming\Python\Python311\site-packages\pymsm\multi_state_competing_risks_model.py:156, in MultiStateModel._assert_valid_input(self)
154 # Check the number of times is either equal or one less than the number of states
155 for obj in self.dataset:
--> 156 n_states = len(obj.states)
157 n_times = len(obj.time_at_each_state)
158 assert n_states == n_times or n_states == n_times + 1

AttributeError: 'str' object has no attribute 'states'

Dear Sir/Madam,

I am confused by the denifition of time_entry_to_origin and time_transition_to_target in some of your examples, especially time_transition_to_target. It seems that time_entry_to_origin can be the the time difference between target state and baseline.

Is time_transition_to_target the time difference between target state and origin state OR the time difference between target state and baseline?

Hi! I am one of the reviewers for your JOSS paper. I will be adding comments as issues in the next hours/days.

I tried following the links to the full examples on the README but they are dead.