nirs4all.pipeline.dagml.process_adapter module

nirs4all process adapter — speaks the dag-ml JSONL frame protocol.

Wraps run_node() in the stdin/stdout control-frame loop that dag-ml-cli drives (--describe handshake, then init/task/close frames, or bare one-shot tasks). Mirrors dag-ml/examples/adapters/sklearn_process_controller.py exactly, but the node execution fetches real SpectroDataset rows + real operators instead of synthesizing.

Launched per worker by the coordinator. main loads the dataset and the compiled graph from env (N4A_DAGML_DATASET_PATH / N4A_DAGML_GRAPH_PATH) — the coordinator passes only sample_ids per task, so the adapter owns the sample_id → X/y materialization (the resolver) and the node_id → operator lookup (the compiled graph). run_jsonl_loop() is dependency-injected so it is testable in-process with no subprocess.

Note: the launcher must re-exec this under the project venv — dag-ml-cli runs a .py adapter under the system python3 (which lacks nirs4all), so a non-.py shim is used.

nirs4all.pipeline.dagml.process_adapter.describe() → dict[str, Any]

The coordinator handshake description (modes + capabilities advertised).

nirs4all.pipeline.dagml.process_adapter.main(argv: list[str]) → int

nirs4all.pipeline.dagml.process_adapter.run_jsonl_loop(infile: IO[str], outfile: _Writer, handle: Callable[[dict[str, Any]], dict[str, Any]]) → None

Drive the control-frame loop: init/close acks, task results, bare one-shot tasks.