Install

Warning

Pre-implementation. This page describes proposed contracts. Behavior is subject to change before code lands.

Install the EMMA (Electromagnetic Multi-task Model Assessment) harness and its data dependency. Until the package ships, the commands below are proposed and not yet runnable; the signatures match API / CLI so an implementer can turn each into a command directly.

Install the package

EMMA is distributed as the emma-benchmark Python package (planned).

$ pip install emma-benchmark

Confirm the CLI (command-line interface) is on your PATH:

$ emma list-tasks --version v0.1

Python and library versions

  • Python 3.10 or newer.

  • PyTorch (torch) 2.0 or newer, for tensors, autograd, and the metric backends.

  • torchmetrics 0.11 or newer, scipy 1.11 or newer, numpy, and pydantic 2.0 or newer for schema validation.

These install automatically as dependencies of emma-benchmark. If you need a specific CUDA build of torch, install it before emma-benchmark so pip does not overwrite it.

The rfgen dependency

Dataset regeneration shells out to rfgen, Superpose’s synthetic RF (radio-frequency) scene generator. The emma download command calls rfgen under the hood; EMMA never generates I/Q (in-phase and quadrature) samples itself.

Install rfgen separately, following its repository, and ensure rfgen is importable or on your PATH before you run emma download. A recipe pins a specific rfgen commit; if your installed rfgen is older than the pinned commit, emma download reports the mismatch and points at the required commit rather than silently producing stale scenes.

Verify the install

$ emma list-tasks --version v0.1

ID            PILLAR         METRIC                      OOD AXIS              STATUS
E-LOC-AOA     localization   mean angular error (deg)    unseen environment    v0.1
E-LOC-LOS     localization   balanced accuracy / AUC     unseen environment    v0.1
E-ID-DRONE    identity       AUC                         unseen environment    v0.1
E-ID-FP       identity       top-1 / EER                 leave-one-unit-out    v0.1
E-ID-AMC      identity       accuracy (continuity)       unseen SNR regime     v0.1

When the task table prints, the install is ready. AUC is area under the ROC (receiver operating characteristic) curve; EER is equal error rate; SNR is signal-to-noise ratio; OOD is out-of-distribution. The full per-task metric list is in Task reference. The next step is Quickstart.

See Also