JABS IO (jabs-io)

This package handles all serialization/deserialization logic.

Overview

jabs-io decouples the JABS data representation from specific file formats or legacy versions. It provides a unified way to interact with pose estimation data, behavioral features, and annotations.

API

The package provides a high level API that's intended to cover most use-cases.

from jabs.io import save, load
from jabs.core.types.keypoints import FrameKeypoints

data_instance = load('frames.json', FrameKeypoints)
save(data_instance, 'frames.parquet')

Installation extras

jabs-io ships optional extras for format-specific backends:

HDF5 is not an extra: h5py is a required dependency, since pose files, prediction files and the default feature cache are all HDF5. pyarrow is a direct dependency of jabs-behavior-classifier, so the parquet backend is available automatically when jabs-io is installed as part of the full JABS application. The nwb extra must always be installed explicitly:

pip install "jabs-behavior-classifier[nwb]"

When using jabs-io as a standalone library:

pip install "jabs-io[nwb,parquet]"

Development

Data models are defined as dataclasses in jabs-core: jabs.core.types. Some backends will be able to implicitly handle most dataclasses, but if the dataclass is complicated, or if there is special handling required of the type for a backend, then a type specific adapter should be defined.

All adapters must inherit from jabs.io.base.Adapter.

For convenience, backend specific subclasses are provided that handle shared functionality for specific file backends.

from jabs.io.base import (
    JSONAdapter,
    ParquetAdapter,
    # TODO: HDF5Adapter,
)

To register your adapter for use, use the register adapter decorator.

from jabs.core.enums import StorageFormat
from jabs.io.base import JSONAdapter
from jabs.io.registry import register_adapter

@register_adapter(StorageFormat.JSON)
class DataclassJSONAdapter(JSONAdapter):
    ...