simuci

🇪🇸 Español: Léeme en español

ICU discrete-event simulation engine — distribution sampling, patient clustering, and statistical validation.

Installation

pip install simuci

For development:

git clone https://github.com/coslatte/simuci.git
cd simuci
pip install -e ".[dev]"

Quick Start

from simuci import Experiment, single_run, multiple_replication

# Create an experiment with patient parameters
exp = Experiment(
    age=55,
    diagnosis_admission1=11,
    diagnosis_admission2=0,
    diagnosis_admission3=0,
    diagnosis_admission4=0,
    apache=20,
    respiratory_insufficiency=5,
    artificial_ventilation=1,
    uti_stay=100,
    vam_time=50,
    preuti_stay_time=10,
    percent=3,
)

# Single replication (requires centroids CSV)
result = single_run(exp, centroids_path="path/to/centroids.csv")
print(result)
# {'Tiempo Pre VAM': 5, 'Tiempo VAM': 89, 'Tiempo Post VAM': 168, 'Estadia UCI': 262, 'Estadia Post UCI': 45}

# Multiple replications → DataFrame
df = multiple_replication(exp, n_reps=200, centroids_path="path/to/centroids.csv")
print(df.describe())

Using Your Own Centroid Data

You must pass the path to your centroid CSV explicitly:

from simuci import single_run, Experiment

exp = Experiment(age=55, ..., validate=False)

# Point to your centroids CSV
result = single_run(exp, centroids_path="path/to/real_centroids.csv")

The centroids CSV must have:

• An index column (cluster IDs: 0, 1, 2)
• At least 11 numeric columns (features used for nearest-centroid classification)

You can also use the loader directly:

from simuci.io.loaders import CentroidLoader

loader = CentroidLoader()
centroids = loader.load("path/to/centroids.csv")  # returns numpy array

Statistical Validation

import numpy as np
from simuci import SimulationMetrics, Wilcoxon, Friedman

# Compare simulation output to real data
metrics = SimulationMetrics(
    true_data=np.array(...),       # (n_patients, n_variables)
    simulation_data=np.array(...), # (n_patients, n_replicates, n_variables)
)
metrics.evaluate(confidence_level=0.95, result_as_dict=True)

print(metrics.coverage_percentage)
print(metrics.error_margin)
print(metrics.kolmogorov_smirnov_result)
print(metrics.anderson_darling_result)

Input Validation

All Experiment inputs are validated on construction by default:

from simuci import Experiment

# This raises ValueError: age must be between 14 and 100
Experiment(age=200, ...)

Skip validation with validate=False if you've already validated externally.

API Reference

Architecture

Project map (core vs. validation, I/O, statistics, tooling): docs/architecture.md

License

MIT