pyannote-pipeline

• BREAKING(cli): switch to latest pyannote.database API

• feat: add "seed" parameter for reproducible optimization

• feat(cli): add "device" section in configuration file

• feat(cli): add "--registry" option for custom database loading

• feat(cli): add "--average-case" option to optimize for average case

• setup: switch to optuna 3.1+

• feat: add support for optuna Journal storage

• BREAKING: optimize loss estimate upper bound instead of average (#42)

• feat: add tests and typing (#41, @hadware)

• feat: add ParamDict structured hyper-parameter (#40, @hadware)

• feat: set sub-pipeline "training" attribute recursively (#39)

• doc: fix various typos

• feat: add Pipeline.instantiated attribute

• feat: add Pipeline.training attribute

• BREAKING: remove "direction" argument from Optimizer

• feat: add Pipeline.get_direction method (defaults to "minimize")

• feat: add progress bar in "apply" mode

• feat: add show_progress option to control second progress bar

• improve: catch optuna's ExperimentalWarning

• feat: add second progress bar to display trial internal progress

• fix: skip Frozen parameters in pipeline.instantiate (@PaulLerner)

• fix: switch to pyannote.database 4.0+ (@PaulLerner)

• setup: switch to optuna 1.4+ and pyannote.core 4.0+

• feat: add "direction" parameter to Optimizer

• fix: fix support for in-memory optimization (when db is None)

• setup: switch to pyannote.database 3.0

• feat: add option to bootstrap optimization with pretrained pipeline

• BREAKING: write "apply" mode output to "train" subdirectory

• feat: store best loss value in "params.yml"

• fix: handle corner case in pyannote.pipeline.blocks.clustering

• fix: use YAML safe loader

• feat: add support for callable preprocessors

• setup: switch to pyannote.core 3.0

• setup: switch to pyannote.database 2.2

• fix: do not raise FileExistsError when output directory exists in pyannote-pipeline apply

• fix: skip evaluation of protocols without groundtruth in `

pyannote-pipeline apply` - setup: switch to pyannote.database 2.1

• feat: add export to RTTM format

• setup: switch to pyannote.database 2.0

• fix: fix "use_threshold" parameter in HAC block

• feat: add support for pyannote.metrics (through Pipeline.get_metric)

• feat: add support for optuna trial pruning

• feat: keep track of processing & evaluation time

• feat: switch to optuna backend

• feat: add "use_threshold" option to HAC pipeline

• BREAKING: update Pipeline API

• BREAKING: update Optimizer API

• BREAKING: remove tensorboard support (for now)

• first public release