tufts-ml / aced-hmm-hospitalized-patient-trajectory-model Goto Github PK

Hi @gvisani here are some notes on ways to clean up this repo / make more useable

1. Might be good to have a "So You Want To Model a Specific Intervention" guide... could describe:

• What does someone need to do to try out several different admissions scenarios
• What does someone need to do to try out different transition probas? Durations?

2. Might be good to have a "So You Want To Make a Forecast on a New Dataset" guide... could be a checklist like:

• Create folder of count data files in needed format
• Decide on these settings (talk about what needs to go into config.json)
• Run forecast using pretrained parameters from CA (show how to run it and interpret results)
• Run ABC to specialize parameters to this specific region (describe what's needed to do that, point to other docs that give details)
• ---- How can they set priors? What should they know about defaults if they don't want to mess with this?
• ---- How can they adjust distance settings? What should they know about defaults if they don't want to mess with this?

3 Useability

Could there be a smarter "default" about warmup, so user doesn't have to decide on specific counts for the 4 previous days? What if we just defaulted to 1/P scaling for each of P previous days, and we suggest users pick P=4 or something. Then no counts need to be provided for anywhere but the daily_XYZ.csv files....

Essentially, instead of current config.json format, which has:

  "init_num_InGeneralWard": {"-4": 75, "-3": 75, "-2": 75, "-1": 75, "0": 79},
  "init_num_OffVentInICU": {"-4": 16, "-3": 16, "-2": 16, "-1": 16, "0": 16},

I'm wondering if we can just have a line like:

  "num_past_timesteps_for_warmup" = 4,
  "scale_factor_for_init_counts_in_warmup_period_before_day0" = {-3: 1/4, -2: 1/4, ... 0: 1/4}

4 Maintenance

• Can we remove PatientTrajectory.py entirely? Seems like that's not needed anymore, right? We use the simulate_traj.... scripts
• Can we remove the Snakemake stuff (from installation and from workflows)? Seems like we didn't really use that in the end
• for the provided datasets, can we remove the first (unnamed) column?

Hello! I have ran your sample workflow, but it seems like to have two small problems in Step 4.
First,it is aced_hmm.ABC_test_metric,not aced_hmm.abc_test_metric,second it should be --config_path instead of --config_file.
It should be:
$ python -m aced_hmm.ABC_test_metrics --input_dir [results/US/MA-20201111-20210111-20210211] \ --output_dir [results/US/MA-20201111-20210111-20210211] \ --output_template [metrics_after_abc] \ --config_path [results/US/MA-20201111-20210111-20210211/config_after_abc.json] \ --true_stats [datasets/US/MA-20201111-20210111-20210211/daily_counts.csv] \ --input_summaries_template [summary_after_abc_] \ --coverages [2.5_97.5,10_90,25_75] \ --comma_sep_expected_columns [n_InGeneralWard,n_OffVentInICU,n_OnVentInICU,n_InICU,n_occupied_beds,n_TERMINAL,n_TERMINAL_5daysSmoothed] .
Thanks!