[Breaking Changes]

[Added]

[Changed]

[Deprecated]

[Removed]

[Added]

• PR #1649: Add progress bar for label model training

[Changed]

• PR #1645: Upgrade networkx requirement to support networkx>=2.5
• PR #1663: Upgrade tensorboard requirement to support tensorboard>=2.0
• PR #1652: Fix issue with logging display during label model training
• PR #1677: Bump version upper bound of networkx to 2.7

[Contributors]

Thanks to @anerirana, @thatch, @sfilipi, @hardianlawi, @asottile and @marekmodry for contributions!

[Breaking Changes]

• PR #1633, #1628: Update requirements for numpy (>=1.16.5,<1.20.0), pandas (>=1.0.0,<2.0.0), and scikit-learn (>=0.20.2,<0.25.0)

[Added]

• PR #1628: Add unit test coverage for Python 3.8
• PR #1616: Accept prec_init as array or list in LabelModel
• PR #1602: Add get_label_instances to analysis module

[Contributors]

Thanks to @DavidKoleczek and @antonis19 for contributions!

[Added]

• PR #1572: Allow specification of memoize_key in preprocessors
• PR #1597: Improved error messages for MultitaskClassifier
• PR #1592: Improved LabelModel documentation

[Contributors]

Thanks to @Wirg, @TrigonaMinima, and @dchichkov for recent contributions!

[Breaking Changes]

• PR #1565: Upgrade pytorch requirement to support torch>=1.2.0

[Breaking Changes]

• PR #1535: Refactor baseline model imports in snorkel.labeling.model
  • Now, from from snorkel.labeling import MajorityLabelVoter, LabelModel can be expressed from snorkel.labeling.model import MajorityLabelVoter, LabelModel

[Added]

• PR #1533: Allow option to save optimizer state in Trainer
• PR #1523: Add GPU option for spaCy preprocessor

[Changed]

• PR #1540: Fix squeeze bug in to_int_label_array function
• PR #1520: Fix error bucket documentation in snorkel.analysis.error_analysis

[Contributors]

Thanks to @rjurney, @ptrcklv for recent contributions!

[Changed]

• PR #1502: Faster symmetry breaking in LabelModel using Munkres algorithm

[Breaking Changes]

• PR #1481: removed fault tolerant mode for labeling functions

[Added]

• PR #1481: fault tolerant mode for appliers

[Changed]

• PR #1450, 1467: ignore abstains in scoring, except coverage
• PR #1463: serialize all attributes of label model
• PR #1466: fix label model GPU training option
• PR #1477, #1492: pin dependency versions

[Removed]

• PR #1454: set_seed utility removed

[Contributors]

Thanks to @HiromuHota, @ferhatelmas, and @garaud for their recent contributions!

[Breaking Changes]

• PR #1453: SlicingClassifier renamed to SliceAwareClassifier

[Added]

• PR #1451: add heuristic for breaking symmetry in multiple label model optima case
• PR #1442: integration test for MultitaskClassifier

[Changed]

• PR #1444: fix label model weight clamping behavior
• PR #1445: fix JSON log writer
• PR #1447: fix correct/incorrect count bug in LFAnalysis
• PR #1428, 1449: catch invalid label model inputs
• PR #1441: make inputs to Scorer.score optional

Version 0.9.0 is a complete redesign of the Snorkel library. There's too much added, changed, and removed to list in this entry. For more information on the release, check out the project homepage. From here forward, we'll keep a detailed changelog.