events¶

CheckpointSaved¶

class CheckpointSaved(model_dir)¶

This event should be emitted when a model is saved as part of checkpointing in a callback.

Parameters: model_dir (str) – the directory the model was saved to.

model_dir: str¶

property msg¶

EarlyStop¶

class EarlyStop(epoch)¶

this event should be emitted when training is instructed to stop early.

Parameters: epoch (int) – the epoch that stopping is triggered.

epoch: int¶

property msg¶

Event¶

class Event¶

base class for all events.

property msg: str¶

Return type: str

logmsg()¶

Return type: str

Improvement¶

class Improvement(field, stage, steps, best, last_best)¶

This event should be emitted on an monitored improvement.

Parameters

field (str) – the field being monitored.
stage (str) – the stage being monitored.
steps (int) – the number of steps (generally epochs) that between this and the last improvement.
best (float) – the best value.
last_best (float) – the last best value

best: float¶

last_best: float¶

property msg: str¶

Return type: str

MonitoringEvent¶

class MonitoringEvent(field, stage, steps)¶

base class for monitoring events

field: str¶

stage: str¶

steps: int¶

logmsg()¶

Return type: str

Stagnation¶

class Stagnation(field, stage, steps, best)¶

This event should be emitted when a monitored metric or loss is not improving.

Parameters

field (str) – the field being monitored.
stage (str) – the stage being monitored.
steps (int) – the number of steps (generally epochs) that the measured value has been stagnant.
best (float) – the best value seen so far.

best: float¶

property msg: str¶

Return type: str

UnbottleEvent¶

class UnbottleEvent(epoch, block, lr)¶

this event should be emitted when fintuning.

Parameters: epoch (int) – the epoch that stopping is triggered.

epoch: int¶

block: str¶

lr: float¶

property msg¶

UnbottlingComplete¶

class UnbottlingComplete¶

this event should be emitted when unbottling is complete during finetuning.

Parameters: epoch – the epoch that stopping is triggered.

property msg¶