Model prediction class probability distribution bugfix

• Fixes a bug where model prediction uploads would error out if the class_pdf field was not set, even though it's optional.

Remove grequests, better error handling

• Remove grequests as a dependency
• Switch to using aiohttp for local file uploads
• Add warning for large uploads
• Local file uploads will now crash loudly if anything goes wrong on any batch instead of silently proceeding with errors
• Disallow sending NaNs for now since that would mess up querying
• Match all dependency requirements with colab so we stop getting errors when trying to use the client in colab

Async annotation ingest + Send to scale

• Support for large batch asynchronous ingest of annotations/groundtruth/segmentations
• Support for sending a slice to labeling project via API

Async Prediction Ingest

• Adds support for ingesting predictions of all types asynchronously in order to unblock large uploads.

Async Dataset Item ingest

• Dataset.append force argument replaced with update (was inconsistent with documentation)
• Async ingest option available (dataset.append now has an asynchronous flag)
• Async Job object added which can poll status
• Error check for ingesting duplicate reference ids

v0.1.3

• Fixed bug leading to confusing errors on trying to create an empty slice.
• Fixed bug with ndjson serialization.

v0.1.2

• Export autotag scores from dataset object
• Better error surfacing for errors from API
• Bugfixes to GCS urls

April 21st Release

• Improve repr for all objects
• Increase client-side timeout to reduce upload flakiness
• Add slice.items()