Input Format
This page describes the required input formats for running the LoS Estimator.
Configuration Files
Configuration is managed via TOML files. The default configuration at los_estimator/default_config.toml provides all necessary settings.
Key Configuration Sections
- data_config
icu_file: Path to ICU occupancy data CSV.los_file: Path to hospital LOS data CSV.start_dayandend_day: Time range for analysis.
- model_config
kernel_width: Width of distribution kernel in days (default: 120).train_width: Width of training window in days (default: 102).test_width: Width of test window in days (default: 21).step: Step size for sliding windows (default: 7).distributions: List of distributions to fit (e.g.,["lognorm", "gaussian", "linear"]).error_fun: Error function for optimization ("mse","mae", etc.).
- debug_config
one_window: Fit only the first window (bool).less_windows: Reduce windows to ~3 for quick testing (bool).less_distros: Test only linear and compartmental (bool).only_linear: Fit only linear models (bool).
- visualization_config
show_figures: Display plots interactively (bool).save_figures: Save plots to disk (bool).figsize: Figure dimensions as[width, height].
- animation_config
show_figures: Display animations interactively (bool).save_figures: Save animations as GIFs (bool).
Data Files
ICU Data (Required)
The primary input is a CSV file containing ICU admission and occupancy time series with the following columns:
date- Date in YYYY-MM-DD formatadmissions- Number of new ICU admissions on that date (integer)occupancy- Total ICU beds occupied on that date (integer)
Example:
date,admissions,occupancy
2020-01-01,5,20
2020-01-02,3,22
2020-01-03,7,25
...
File Path: Specify the path to this file in your configuration as data_config.icu_file.
Initial Parameters (Optional)
You can provide initial parameter values for optimization to improve convergence. The file should be CSV formatted:
distro- Distribution name (e.g., “lognorm”, “gamma”, “gaussian”)params- Space-separated parameter values in brackets, e.g.,[2.5 1.0 1.2]
Example:
distro,params
lognorm,[2.5 0.8 1.0]
gamma,[3.0 0.5 1.0]
File Path: Specify as data_config.init_params_file in your configuration.
Sample LoS Distribution (Optional)
For validation purposes, you can provide a known LoS distribution (typically for synthetic data). The file should be CSV formatted:
day- Length of stay in days (integer, 0 to max_los)probability- Discharge probability for that day (float, 0-1)
Example:
day,probability
0,0.05
1,0.10
2,0.15
...
File Path: Specify as data_config.los_file in your configuration.
Note: The package includes a sample distribution file for the synthetic example.