pyjpm 0.8.8

Joint Progression Modeling (JPM): A Probabilistic Framework for Mixed-Pathology Progression

pyjpm

Installation

pip install pyjpm

Data generation

To run gen.py first. Then to generate partial ordering datasets. gen_partial.py.

Change Log

• 2025-07-16 (V 0.0.6)

  • Added mp_method = BT in algorithm.py and run.py.
  • Added PL.
  • Fixed the bug in generate_data.py.

• 2025-07-19 (V 0.0.7)

  • Added the class of PlackettLuce.

• 2025-07-20 (V 0.0.15)

  • Updated the definition and implementation of conflict and certainty.
  • Made sure data folder exists after uploading to pypi.
  • Made sure fixed_biomarker_order = True if we use mixed patholgy in data generation.
  • Fixed a bug in the calculation of conflict.
  • Made sure the algorithm.py is using the correct energy calculation functions.
  • Added entropy and certainty calculation in MCMC sampler.
  • Made sure in generate_data.py, certainty is calculated based upon the mp_method.
  • Made sure in generate_data.py, we can tweak the paramter of mcmc_iterations, otherwise it will super slow. This is because the time complexity is mcmc_iterations * sample_count.
  • Tested obtaining ordering_array from separate disease data files. Made some modifcations in algorithm.py to allow this.

• 2025-07-23 (V 0.0.16 -- didn't push to Pypi)

  • Implemented the conflict version of using only discordant pairs.

• 2025-07-25 (V 0.0.16)

  • Updated algorithm.py to reflect changes in the class of PlackettLuce.

• 2025-07-26 (V 0.0.17)

  • Updated generate_data.py to skip calculating certainty and conflict if sample_count <= 1.

• 2025-08-04 (V 0.1.7)

  • Use fastsaebm codes.
  • Finished testing and data generation.
  • With m_{variant} when number of repitition is 1.
  • Fixed overflow bug in prob_accept=min(1.0, np.exp(current_energy - new_energy)).

• 2025-08-04 (V 0.2.9)

  • Implemented the new certainty measure.
  • Used the same rng all throughout in generate data.
  • Added save_details to run.py.
  • Solved the logic bug of save_details and save_results.
  • Ensured the randomness again in generate_data.py.

• 2025-08-09 (V 0.3.1)

  • Used 15 biomarkers.
  • Dynamically adjust dirichlet multinomial alpha array based on the number of biomarkers.

• 2025-08-10 (V 0.3.3)

  • Use numpy and numba (whenever possible) in mp_utils.py.

• 2025-08-11 (V 0.3.6)

  • Updated numba version

• 2025-08-12 (V 0.3.9)

  • Kept improving the numba version. Now it's faster.
  • Include MCMC in PL sampling as well.

• 2025-08-11 (V 0.4.0)

  • Add RMJ distance mallows.

• 2025-08-16 (V 0.4.2)

  • Try all njit in mp_utils.py. I want to test it on CHTC.

• 2025-08-17 (V 0.4.4)

  • I know using np.random is not helpful in shuffle_order func. Change back to the slow version.

• 2025-08-18 (V 0.4.10)

  • Try to use rng in func of obtain_affected_and_non_clusters.
  • In mallows, use BT for central ranking sampling.
  • Added theta = 100 in mallows.
  • Added mp_method = 'random' in generate_data.py.
  • Remobed recodes folder when running mp-ebm.

• 2025-08-19 (V 0.4.19)

  • Corrected an error: in data generation, for experiment 9, the noise_std should be max_length * noise_std_parameter rather than its square root. This is imprtant because after using square root, the noise_std in fact become larger, not smaller. For example, in our example where N = 10, the noise_std should be N*0.05 = 0.5, but after square root, it becomes 0.7. If N = 4, then std should be 02, but becomes 0.45 after square root.
  • Added the temperature parameter for mallows sampling.
  • Reuse params inferred from individual diseases.
  • Used biomarkers and theta/phi params obtained from NACC data analysis.

• 2025-08-20 (v 0.5.1)

  • Used biomarkers and their theta/phi from both ADNI only.
  • Changed 'random' to 'Random' in generate_data.py.
  • Randomly choose two floats for new theta params for overlapping biomarkers.

• 2025-08-21 (V 0.5.4)

  • Try only 12 biomarkers for params.
  • Try 18 biomarkers for params.
  • Add rnadom pertubations to overlapped biomarkers params.

• 2025-08-22 (V 0.5.6)

  • Try scaling factor for energy in mh.py

• 2025-08-23 (V 0.5.9)

  • Test differetn energy influence.
  • use_scaling

• 205-08-25 (V 0.6.0)

  • Test percentile.

• 205-08-26 (V 0.6.2)

  • Added analysis about alignment and effect_size.

• 2025-08-27 (V 0.6.7)

  • Added energy_prior and model_prior.
  • Mapped energy_prior to mallows_temperature.

• 2025-08-28 (V 0.6.9)

  • Removed energy_prior. Only use model calibration.

• 2025-08-32 (V 0.7.3)

  • Modified the data generation just like in subtypes.

• 2025-09-01 (V 0.7.4)

  • Removed the forcing range of event times.

• 2025-09-03 (V 0.7.5)

  • Added save_data boolen to generate_data.py.

• 2025-09-04 (V 0.8.0)

  • Removed calibration. We cannot use it.
  • Aligned with how i get staging with pysaebm: completely blind, not even using healthy ratio and the learned stage prior. Only use the theta/phi.
  • Modified what to return in run.py.

• 2025-09-09 (V 0.8.1)

  • Added plots back.

• 2025-09-21 (V 0.8.2)

  • Removed iteration >= burn_in when updating best_*.

• 2025-10-08 (V 0.8.4)

  • Used soft counts for conjugate prior updates.

• 2025-10-09 (V 0.8.6)

  • Update the non-normal distribution parameters.

• 2025-11-09 (V 0.8.8)

  • Changed the pkg name from pympebm to pyjpm.