uv add qe-tma --index qe-libs=https://qe-libs.org/py/simple/

pip install --extra-index-url https://qe-libs.org/py/simple/ qe-tma

--extra-index-url https://qe-libs.org/py/simple/
qe-tma

pip install --extra-index-url https://qe-libs.org/py/dev/simple/ qe-tma

uv add qe-tma --index qe-libs-dev=https://qe-libs.org/py/dev/simple/

qe-tma

Python Transmodal Analysis (TMA) — sister package to the tma R package (import tma). Formerly pyTMA (import pytma), which was never published.

The math runs in C++ shared with R: accumulation (stanza windows, context tensors, row connections, weight models) is libtma, compiled into this package (tma.libtma); node positions come from qe-ena (libena) and generic numerics from qe-lib (libqe).

Installation

pip install qe-tma --extra-index-url https://qe-libs.org/py/simple/

Development builds from main are on https://qe-libs.org/py/dev/simple/.

Development install

Building from a checkout compiles the libtma extension: it needs a C++17 compiler, CMake and Armadillo (brew install armadillo, apt install libarmadillo-dev), plus the headers that scripts/sync-headers.sh vendors into python/include/ (libtma from this repo, libqe from Conan; set LIBQE_INCLUDE=<libqe checkout>/include to use a local libqe instead).

sh scripts/sync-headers.sh                 # from the repo root
pip install qe-lib qe-ena --extra-index-url https://qe-libs.org/py/simple/
pip install "./python[dev]"
pytest python/tests --import-mode=importlib   # tests the installed package

Quick start

import pandas as pd
from tma import TMA, contexts, context_tensor

# rules / contexts / context_tensor mirror tma's R functions; TMA().fit()
# runs contexts -> accumulate -> normalize -> SVD -> node positions.

The low-level accumulation functions (apply_tensor_unit, aggregate_row_connections, accumulate_stanza, ...) are in tma.libtma.