This will incorporate DLPipelines.jl directly into FastAI.jl and drop the dependency. This is a BREAKING change.
In the process, the following changes will be made to the LearningMethod interfaces:
[x] Rename LearningMethod to LearningTask, method to task
In an earlier version of DLPipelines.jl, LearningTask was an abstract parametric type for LearningMethods to signify that there can be multiple ways of solving a task, but this was dropped at some point.
LearningTask is imo a more intention-revealing name, as method is a very ambiguous term (especially in Julia-land). This is especially true for the short variable name task vs method.
As a result of this, subtypes of LearningMethod will also be renamed. Specifically, [Abstract]BlockMethod -> [Abstract]BlockTask and SupervisedMethod -> SupervisedTask.
methoddataset -> taskdataset and methoddataloaders -> taskdataloaders
findlearningmethods -> findlearningtasks
[x] Clean up the basic LearningMethod interface:
rename encode(method, ctx, sample) to encodesample. This way, encode is reserved for Encodings.
rename decodeŷ to decodeypred . Easier to write and to differentiate from decodey . Old name will be aliased.
This will incorporate DLPipelines.jl directly into FastAI.jl and drop the dependency. This is a BREAKING change.
In the process, the following changes will be made to the
LearningMethod
interfaces:LearningMethod
toLearningTask
,method
totask
LearningTask
was an abstract parametric type forLearningMethod
s to signify that there can be multiple ways of solving a task, but this was dropped at some point.LearningTask
is imo a more intention-revealing name, as method is a very ambiguous term (especially in Julia-land). This is especially true for the short variable nametask
vsmethod
.LearningMethod
will also be renamed. Specifically,[Abstract]BlockMethod -> [Abstract]BlockTask
andSupervisedMethod -> SupervisedTask
.methoddataset -> taskdataset
andmethoddataloaders -> taskdataloaders
findlearningmethods -> findlearningtasks
LearningMethod
interface:encode(method, ctx, sample)
toencodesample
. This way,encode
is reserved forEncoding
s.decodeŷ
todecodeypred
. Easier to write and to differentiate fromdecodey
. Old name will be aliased.