Skip to contents

Version 0.9

iRfcb 0.9.0

CRAN release: 2026-06-08

New features

  • Added ifcb_extract_features(), which computes the slim feature set (version 4) and blob masks from raw IFCB data by calling the WHOI ifcb-features Python package. Features (<bin>_features_v4.csv) and blobs (<bin>_blobs_v4.zip) are written to separate, user-specified folders, existing outputs are skipped unless overwrite = TRUE, and bins can be processed in parallel via parallel = TRUE / n_cores. A cli progress bar advances as each bin is processed (for both sequential and parallel runs), and interrupting the function (e.g. ESC / Stop) reliably terminates the parallel worker processes instead of leaving them writing files in the background.
  • Added features and features_ref arguments to ifcb_py_install() to optionally install the WHOI ifcb-features package (and its dependencies) from GitHub, as required by ifcb_extract_features(). By default the latest published release is installed; features_ref can pin a specific tag or install the development branch. When installing into an existing virtual environment, the install is skipped if ifcb-features already imports successfully (unless features_ref is supplied), avoiding a slow repeated download.
  • Added a new dataset_name argument to ifcb_list_dashboard_bins() to optionally restrict the listing to bins from a specific dataset. This argument remains useful for self-hosted dashboard instances that have not yet updated to remove the api/list_bins endpoint.
  • Added support for the IRFCB_PYTHON_VENV environment variable. When USE_IRFCB_PYTHON = "TRUE", you can now set IRFCB_PYTHON_VENV to either a named virtualenv or a full path to a venv directory to control which Python environment is activated on package load. If unset, the previous behavior of auto-discovering a venv named iRfcb is retained.

Deprecations

Minor improvements and fixes

  • Additional Python packages installed into an existing virtual environment by ifcb_py_install() are now installed with a clean dependency resolution (no longer using pip --ignore-installed). Previously, installing packages with pinned, compiled dependencies (such as ifcb-features/pyifcb, which pin exact numpy/scipy/pandas versions) could layer incompatible builds on top of existing ones and corrupt the environment (e.g. ImportError: cannot import name '_spropack').
  • The default gradio_url for ifcb_classify_images(), ifcb_classify_sample(), ifcb_classify_models(), and ifcb_save_classification() has changed from the Hugging Face example Space (https://irfcb-classify.hf.space) to a more stable instance hosted on SciLifeLab Serve (https://ifcb.serve.scilifelab.se). The default model_name has correspondingly been updated to "SMHI NIVA SYKE SAMS SZN ResNet 50 V6". The Hugging Face Space remains documented as a free alternative for testing and demonstration.
  • Migrated all user-facing messaging from base R (stop(), warning(), message()) and utils::txtProgressBar to the cli package. Errors, warnings, and informational messages now use semantic inline markup (file paths, argument names, function names, values) and pluralization. Progress bars are rendered via cli::cli_progress_bar(). cli is now an Imports dependency.