singhlab 0.0.2

To install:

git clone git@github.com:rohitsinghlab/singhlab-main.git
cd singhlab-main
pip install .

To run sceodesic with 10 different random seeds

from singhlab.utils.decorators import run_with_seed
from sceodesic import run_sceo

seeds = [123, 444, 33, 1234, 99, 14, 37]
results = {}

adata = sc.read_h5ad('adata.h5ad')

for seed in seeds:
    run_with_seed(seed)(run_sceo)(adata, num_hvgs=300)
    sceo_embeddings = adata.obsm['sceo_embeddings']
    sceo_programs = adata.varm['sceo_programs]
    results[seed] = (sceo_embeddings, sceo_programs)

To do a bootstrap for some uncertainty quantification:

from singhlab.utils.decorators import bootstrap
import numpy as np

# bootstrap parameters 
num_trials = 100
seed = 42

# the function we want to run our bootstrap on
@bootstrap(num_trials, seed)
def get_mean(data):
    return np.mean(data)
 
# Example data
data = np.array([5, 7, 8, 6, 9, 10, 12, 7, 6, 8])

# automatically applies bootstrap
boot_means = get_mean(data)

# Percentile CI (95%)
lower, upper = np.percentile(boot_means, [2.5, 97.5])

print(f"Bootstrap mean CI: [{lower:.2f}, {upper:.2f}]")