Skip to main content

Analysis Pipelines

Pipelines connect compatible LgoPy blocks into a repeatable processing workflow. Use them to transform raw images into structured features, run an image correction, or produce measurements for downstream analysis.

Analysis pipelines

Build and run a pipeline​

  1. Inspect available methods in Analysis Modules.
  2. Open Analysis Pipelines and select Run pipeline.
  3. In Dataset & Functions, choose the project, study, dataset, blocks, and versions.
  4. In Functions Ordering, arrange the processing steps.
  5. In Parameters, configure each block for your data.
  6. In Pipeline, review the graph and use Preview JSON to inspect the definition.
  7. Select Run pipeline, then follow the job's progress and output log.

The backend queues runs as durable operations. A configured embedded worker or Celery worker executes them; submitting a run does not mean it has completed.

Match inputs and outputs​

Each block has its own contract. A block may return a table, a dataset item, or a status summary while saving images separately. Connect steps only when the next block can consume the previous output. The validation endpoint POST /api/pipelines/validate and MCP's validate_pipeline_steps check catalog blocks and declared type compatibility without starting a job. They do not establish that a method is scientifically appropriate for your data.

For example, the bundled ndvi_index block returns measurements for multispectral assets. Its nir_band and red_band settings are one-based band indices; check your sensor's band order before running it.

API payload​

A single-block run can use this request body. Replace the dataset ID and verify the installed block version and band indices:

{
  "dataset_id": 7,
  "pipeline_name": "NDVI plot measurements",
  "pipeline": [
    {
      "block": "ndvi_index",
      "version": "0.1.0",
      "args": {"nir_band": 5, "red_band": 3}
    }
  ]
}

Submit to POST /api/pipelines, /api/pipelines/run, or the dataset-scoped /api/pipelines/datasets/7 route. A successful submission returns HTTP 202 and an operation record. Its id is the operation ID; the pipeline job ID is in payload.pipeline_id. Use the pipeline ID with GET /api/pipelines/{pipeline_id} to inspect the run.

Review, rerun, and cancel​

The pipeline page provides output logs, artifacts, rerun, cancel, and delete actions when applicable. A rerun uses the stored pipeline definition. Review failures before retrying, and keep the relevant block versions available. Deleting a pipeline queues removal of its registered artifacts, so download results you need to retain before deleting the run.

Continue with Running a Workflow, Results and Exports, or Analyze Data with the Agent.