Pipelines

A spec names the recordings, the operations and the output. It’s data, so the same spec runs from Python, the command line and the app, and it’s saved beside each result.

JSON

{
  "dspk_pipeline": 1,
  "nodes": {
    "raw": {"op": "read", "path": "recording.bin"},
    "hp": {"op": "iir", "band": "highpass", "cutoff": [300.0], "mode": "forward"},
    "out": {"op": "write", "path": "filtered.zarr", "format": "zarr"}
  },
  "edges": [["raw", "hp"], ["hp", "out"]]
}

TOML

dspk_pipeline = 1
edges = [["raw", "hp"], ["hp", "out"]]

[nodes.raw]
op = "read"
path = "recording.bin"

[nodes.hp]
op = "iir"
band = "highpass"
cutoff = [300.0]
mode = "forward"

[nodes.out]
op = "write"
path = "filtered.zarr"
format = "zarr"

import dspk

dspk.pipeline.plan("spec.toml")             # what it would do, without touching the device
reports = dspk.pipeline.run("spec.toml")

Progress and cancelling

def progress(input, inputs, done, windows):
    print(f"file {input + 1}/{inputs}: {done}/{windows}")
    # return False to cancel

dspk.pipeline.run("spec.toml", progress=progress)

Cancelling (or Ctrl+C) raises KeyboardInterrupt, removes the unfinished output and keeps inputs already finished.

More on specs: Pipelines and the spec reference.