pythonDEA

pythonDEA is a focused Python library for slack-based DEA, environmental frontiers, and panel productivity analysis. It uses NumPy and SciPy for the numerical core and exposes a consistent fit() interface across its models.

Why pythonDEA?

Python already has capable DEA libraries, but lightweight packages tend to focus on classical radial models.

Tools that combine SBM, undesirable outputs, and productivity analysis often target older Python versions or bring desktop interfaces and larger solver stacks.

pythonDEA provides a focused MIT-licensed option for Python 3.10 and newer, with explicit model choices and result objects that retain solver details, slacks, targets, peer information, and reproducibility metadata.

Install

Install the current version from GitHub:

python -m pip install "git+https://github.com/randomcat4/pythonDEA.git"

The core package depends only on NumPy and SciPy. Install the optional tables extra if you want pandas conversion helpers:

git clone https://github.com/randomcat4/pythonDEA.git
cd pythonDEA
python -m pip install -e ".[tables]"

Quick start

The bundled emissions dataset contains three decision-making units with the same input and desirable output but different undesirable output levels.

from pythondea import audit_result, fit
from pythondea.datasets import load_emissions_cross_section

data = load_emissions_cross_section()
result = fit("sbm", data, orientation="bad_output_adjusted")

for row in result.table("efficiency").rows:
    print(f"{row['dmu']}: {row['score']:.6f}")

print("audit passed:", audit_result(result).passed)

Expected output:

clean: 1.000000
balanced: 0.923077
dirty: 0.785714
audit passed: True

audit_result() checks that the result contains its primary table, solver metadata, package version, and a deterministic result hash. It is a metadata check, separate from validating the assumptions of a chosen DEA model.

Use your own data

Rows represent decision-making units and columns represent variables:

from pythondea import DEAData, fit

data = DEAData(
    inputs=[[1.0], [1.0], [1.0]],
    good_outputs=[[1.0], [1.0], [1.0]],
    bad_outputs=[[1.0], [1.2], [2.2]],
    dmu_names=["clean", "balanced", "dirty"],
    input_names=["capital"],
    good_output_names=["output"],
    bad_output_names=["emissions"],
)

result = fit("directional_distance", data, direction="bad_output")

Use PanelDEAData.from_3d() for balanced panel data. The first dimension is period, the second is entity, and the third is variable. DataFrame adapters are available as dea_from_dataframe() and panel_from_dataframe() when the tables extra is installed.

Available models

Model name	Purpose	Data type
sbm	Slack-based efficiency with CRS or VRS, orientation choices, and undesirable outputs	DEAData
sbm_super_efficiency	Exclude-self SBM for ranking efficient units and checking frontier sensitivity	DEAData
sbm_malmquist	Adjacent-period SBM-Malmquist productivity decomposition	PanelDEAData
directional_distance	Directional distance analysis for environmental frontiers	DEAData
malmquist_luenberger	Adjacent-period green productivity analysis based on directional distances	PanelDEAData

Run list_models() to inspect the registered model names. model_catalog() also returns each model's family, summary, keywords, and citation hint.

Work with results

Every model returns a ModelResult with the same basic interface:

primary = result.table()
rows = primary.rows
frame = primary.to_pandas()  # requires the tables extra

json_text = result.to_json()
run_hash = result.reproducibility_hash()
solver = result.metadata["solver_backend"]

SBM results include efficiency, slack, and target tables. Directional distance results include distance and target tables. Panel models return one row per entity and adjacent-period transition.

Research scope

pythonDEA concentrates on non-radial and directional methods used in environmental efficiency and panel productivity research. Current development focuses on numerical validation and richer frontier diagnostics. The model registry also lets external estimators use the same data and result interfaces.

Repository layout

Path	Role
v1/	Original standalone SBM-Malmquist tool and its archived documentation
v2/	Tested numerical core for data, frontiers, SBM, DDF, and panel calculations
v3/pythondea/	Current importable package; the directory name is retained from the package-layer rewrite
examples/	Small scripts for SBM-Malmquist and green productivity reproduction
tests/	Numerical, API, adapter, audit, and example tests

The package version is independent of the historical directory names. The current release metadata reports version 4.0.0.

Development

git clone https://github.com/randomcat4/pythonDEA.git
cd pythonDEA
python -m pip install -e ".[dev,tables]"
python -m pytest

The repository also includes two runnable examples:

python examples/v3_sbm_malmquist_reproduction.py
python examples/v4_green_productivity_reproduction.py

The staged implementation plan is in TODO.md.

License

pythonDEA is released under the MIT License.