FluxTraining
"""
module Events
Provides the abstract [`Event`](#) type and concrete event types.
Events in [`TrainingPhase`](#) and [`ValidationPhase`](#):
- [`EpochBegin`](#) and [`EpochEnd`](#), called at the beginning and end of each
epoch.
- [`StepBegin`](#) and [`StepEnd`](#), called at the beginning and end of each
batch.
- [`LossBegin`](#), called after the forward pass but before the loss calculation.
[`TrainingPhase`](#) only:
- [`BackwardBegin`](#), called after forward pass and loss calculation but before gradient
calculation.
- [`BackwardEnd`](#), called after gradient calculation but before parameter update.
"""
module
Events
"""
abstract type Event
Abstract type for events that callbacks can hook into
"""
abstract
type
Event
end
"""
EpochBegin()
[`Event`](#) called at the beginning of an epoch.
"""
struct
EpochBegin
<:
Event
end
"""
EpochEnd()
[`Event`](#) called at the end of an epoch.
"""
struct
EpochEnd
<:
Event
end
"""
StepBegin()
[`Event`](#) called at the beginning of a batch.
"""
struct
StepBegin
<:
Event
end
"""
StepEnd()
[`Event`](#) called at the end of a batch.
"""
struct
StepEnd
<:
Event
end
"""
LossBegin()
[`Event`](#) called between calculating `y_pred` and calculating loss
"""
struct
LossBegin
<:
Event
end
"""
BackwardBegin()
[`Event`](#) called between calculating loss and calculating gradients
"""
struct
BackwardBegin
<:
Event
end
"""
BackwardEnd()
[`Event`](#) called between calculating gradients and updating parameters.
"""
struct
BackwardEnd
<:
Event
end
export
# abstract
Event
,
# concrete
EpochBegin
,
EpochEnd
,
StepBegin
,
StepEnd
,
LossBegin
,
BackwardBegin
,
BackwardEnd
endmodule
using
.
Events