haMSM error while construction of flux matrix

[Smriti P]

Hi everyone,

I'm trying to implement the haMSM westpa plugin for my system, but while constructing thr flux matrix my runs crash with the error "Still not clean after cleaning!". What could be the possibel sources for this error? Here are the details of the extension part in my configuration file:

 plugins:
    - plugin: westpa.westext.hamsm_restarting.restart_driver.RestartDriver
      n_restarts: 0
      extension_iters: 0
      n_runs: 1
      coord_len: 2
      n_restarts_to_use: -1
      initialization_file: restart_initialization.json
      ref_pdb_file: commonfiles/ABCD_unbound_neutral.pdb
      model_name: twodimersflux
      n_clusters: 10
      we_folder: .
      basis_pcoord_bounds: [[51.0,60.0]]
      target_pcoord_bounds: [[0,14.2]]
      pcoord_ndim0: 1
      dim_reduce_method: none
      parent_traj_filename: parent.dcd
      child_traj_filename: seg.dcd
      user_functions: westpa_scripts/restart_overrides.py
      struct_filetype: mdtraj.formats.PDBTrajectoryFile
      n_cpus: 3

Please let me know if any additional information is needed.

Best,

Smriti Pradhan

[Jeremy Leung]

Hello Smriti,

The "still not clean after cleaning" suggests that the clusters/microbins were not well chosen, so subsequent steps on cleaning the flux matrix is not successful because there are still disconnected sets. You may be able to see more information by adding in `debug: True` into the plugin block.

The haMSM restarting plugin uses your WE bins for the stratified clustering, so you might want to redefine your bin boundaries or change the number of clusters to see if that results in better clusters. (Backup your west.h5 and run` w_bins rebin` and/or `w_truncate` on your last iteration to rebin) Alternatively, you can use your augmented west.h5 file to run the whole haMSM analysis in a jupyter notebook (see here), and come back to the restarting plugin once you found optimal parameters for building the haMSM. You can specify a custom bin mapper by passing in a bin mapper to the clustering step as cluster_args, which will speed up the process.

Hope that helps.

Best,

Jeremy L.