sepa-kim 0.1.0

Segment Profile Extraction via Pattern Analysis

sepa

Segment Profile Extraction via Pattern Analysis

Python port of the R SEPA package (CRAN, Kim 2026).

---

Overview

SEPA implements SVD-based row-isometric biplot analysis for multivariate profile data. After ipsatization (row-mean centering), SEPA provides:

• Plane-fit rho and direction cosines for every person × domain pair
• BCa bootstrap CIs for domain loading coordinates
• Percentile bootstrap CIs for per-person cosines
• Parallel analysis for dimensionality selection

Reference: Kim & Grochowalski (2019) doi:10.1007/s00357-018-9277-7

---

Installation

pip install sepa

Requirements: Python >= 3.9, numpy >= 1.24, scipy >= 1.10

---

Quick start

from sepa import run_sepa, fake_wj, SEPA_DOMAINS

# Full pipeline on the built-in WJ-IV look-alike dataset (n=5127)
res = run_sepa(
    fake_wj,
    K=4,
    domain_names=SEPA_DOMAINS,
    target_ids=[724, 944],   # persons for per-person cosine CIs
    B_dom=2000, B_cos=2000,
    pa_B=2000, seed=1,
)

print(res["pa"]["sig_dims"])           # significant dimensions
print(res["sepa_stats"]["rho"].head()) # plane-fit rho
print(res["dom_coords"])               # domain coords with BCa CIs
print(res["cosine_tables"][724])       # person 724 cosine table

---

Lower-level functions

from sepa import (
    parallel_analysis_ipsatized,  # PA for ipsatized data
    sepa_stats_all,               # rho + direction cosines
    percentile_ci_mat,            # percentile CIs from bootstrap matrix
    boot_cis_all,                 # BCa CIs for all statistics
)

---

Built-in dataset

fake_wj is a 5127 x 8 synthetic WJ-IV look-alike dataset (ID + 7 domains): LT, ST, CP, AP, VP, CK, FR — all in standard score metric (mean100, SD15).

from sepa import fake_wj, load_fake_wj, SEPA_DOMAINS
df = load_fake_wj()   # pandas DataFrame version

---

R package

CRAN: https://cran.r-project.org/package=SEPA GitHub: https://github.com/sekangakim/SEPA

---

Citation

Kim, S.-K., & Grochowalski, J. H. (2019). Segment Profile Extraction via Pattern Analysis. doi:10.1007/s00357-018-9277-7

---

License

MIT

Author

Se-Kang Kim, Ph.D. Baylor College of Medicine / Texas Children's Hospital [e-mail]