Comparing RS — Python option

Runs a full surrogate MDS comparison between two choice datasets and returns a p-value: are the perceptual maps for two datasets genuinely different, or could the difference be explained by chance?

See Installation to set this up.

from rs_tools.compare import compare
from src.rs_py.utils.util import load_choices

# load two datasets
resp1, rep1, _, stims1 = load_choices("choices_A.mat")
resp2, rep2, _, stims2 = load_choices("choices_B.mat")

result = compare(resp1, rep1, stims1, resp2, rep2, stims2,
                 dim=3, n_surrogates=100)

print("p-value:", result['p_value'])
print("Real disparity:", result['real_disparity'])
print("Surrogate mean:", result['surrogate_mean'])

Output:

• Log-likelihood and Procrustes disparity for each dataset
• Null distribution of surrogate disparities vs. the real disparity
• p-value: fraction of surrogates with disparity ≥ real

Resampling methods

Method	Description
with_replacement	Draw observations with replacement from pooled data (default)
without_replacement	Draw without replacement from pooled data
without_replacement_paired	Draw without replacement, keeping A and B draws non-overlapping

result = compare(resp1, rep1, stims1, resp2, rep2, stims2,
                 dim=3, n_surrogates=100,
                 method='without_replacement_paired')

Random seed control

By default each surrogate seeds deterministically from its own index, so re-running with the same n_surrogates reproduces the same surrogate set. For the real-data fit specifically, control the starting point with rng_control.initialize_random_state(if_frozen) before calling compare — see Comparing RS → Python tools UI for what each if_frozen value means.

For the full function reference, see Python function index.