Quickstart

This short example follows the main DEAPack workflow: prepare a table, fit one DEA model, and inspect both the overall results and one organization’s benchmark. Run the three code blocks in order.

1. Prepare the data

from deapack import BCCInput, DEAData, load_dataset

frame = load_dataset("slacks_2x2")
print(frame.to_string(index=False))

data = DEAData.from_frame(
    frame,
    dmu="dmu",
    inputs=("labor", "capital"),
    outputs=("service", "quality"),
)

Each row is a decision-making unit (DMU). DEAData tells DEAPack which column identifies the DMU, which resources should be reduced, and which services should be protected or expanded. For your own study, replace frame and the column names with those from your pandas DataFrame. See the data guide for panel layouts, variable roles, and validation, or the dataset guide for the bundled examples.

2. Fit the model

model = BCCInput()
result = model.fit(data)

BCCInput asks how much each DMU could proportionally reduce its inputs while protecting its outputs, using a variable-returns-to-scale benchmark. See the radial DEA guide for input versus output orientation and CRS versus VRS. If your research question requires another model family, use the method catalog.

3. Check and interpret the result

summary = result.summary()
summary_columns = [
    "dmu_id", "efficiency", "is_efficient",
    "max_slack", "score_valid", "solver_status",
]
print(summary[summary_columns].round(3))

focus = "E"
status = summary.set_index("dmu_id").loc[focus]
assert status[["score_valid", "target_valid", "peer_valid"]].all()

target_columns = ["role", "variable", "observed", "target"]
peer_columns = ["reference_dmu_id", "lambda"]
print("\nTarget for E:")
print(result.targets_for(focus)[target_columns].round(3))
print("\nReference organizations for E:")
print(result.peers(focus)[peer_columns].round(3))

Check score_valid and solver_status before interpreting a score. Here E’s efficiency is about 0.753, indicating a proportional input-reduction opportunity of about 24.7% under this model. Its positive max_slack shows that the radial score is not the whole improvement path. The target records the fitted input levels and remaining output improvements, while the peer weights reconstruct that benchmark from B and C. These are conditional benchmarking results, not causal findings or automatic management instructions.

The result guide explains scores, validity, targets, slacks, and diagnostics in depth; the reference-set guide explains peer weights and alternate benchmarks. From the same result, continue to visualization or reporting and export when needed.