pysemforge

Structural Equation Modeling for Python — closing the R-Python gap for behavioral and econometric modeling.

A Python-native SEM/CFA package inspired by lavaan (syntax) and semTools (multigroup + measurement invariance), with a graphviz-free path diagram backend.

Why pysemforge?

Python has mature estimation backends (e.g. semopy), but still lacks a single package that combines:

The two highlighted rows are the real gaps this package targets — so you stay in Python end-to-end.

Quick Start

import pandas as pd
from pysemforge import Model

# 1) lavaan-compatible syntax
syntax = """
    visual =~ x1 + x2 + x3
    textual =~ x4 + x5 + x6
    speed =~ x7 + x8 + x9
"""

model = Model(syntax)
result = model.fit(data)          # a pandas DataFrame of your indicators
print(result.summary())

# 2) Path diagram (no graphviz needed)
model.plot("sem_path.png", standardized=True)

# 3) Multigroup + measurement invariance (the headline feature)
inv = model.measurement_invariance(data, group_col="gender")
print(inv.summary())
inv.plot("invariance.png")

Install

pip install pysemforge

For the latest development version:

git clone https://github.com/your-username/pysemforge.git
cd pysemforge
pip install -e .

Features

• lavaan-compatible syntax — =~, ~, ~~, ~1, ==, comments, group blocks
• CFA / structural regression / mediation as first-class models
• Estimators — ML, MLR (robust SE), WLSMV (polychoric for ordinal data)
• Native multigroup analysis with equality constraints
• Measurement invariance — fully automated configural → metric → scalar ladder with Δχ² / ΔCFI / ΔRMSEA decisions (R semTools parity)
• Partial invariance — build-up / tear-down of non-invariant parameters
• Fit indices — χ², CFI, TLI, RMSEA (+90% CI), SRMR
• graphviz-free diagrams — publication-quality path plots via matplotlib (no system dependencies)
• pandas / NumPy / SciPy first-class interoperability

Measurement Invariance Example

inv = model.measurement_invariance(data, group_col="group")

# Access each level's fit
inv.levels["configural"]
inv.levels["metric"]
inv.levels["scalar"]

# Pairwise comparisons (Δχ², ΔCFI, decision)
inv.comparisons

The returned object mirrors semTools::measurementInvariance() output, so R users feel at home.

Documentation & Publishing

• Source: https://github.com/your-username/pysemforge
• Docs: https://pysemforge.readthedocs.io
• Issues: https://github.com/your-username/pysemforge/issues

Publish to PyPI (main release target):

pip install build twine
python -m build
twine check dist/*
twine upload --repository testpypi dist/*   # test first
twine upload dist/*                         # then production

Roadmap (post-MVP)

Latent growth curve models · Bayesian estimation · complex survey designs · full lavaanPlot-parity plotting.

License

MIT — see LICENSE.