You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Finally, the interface for the Transform class could be cleaned up, particularly when it comes to having to define a nested function in the transform method.
then there is no need for transform(), and __call__() would handle the call to _apply(), and _apply() would have access to _transform() directly instead of needing to pass it as an argument.
Note: Would need to think of a way to handle transformations that require all of the observation sequences rather than just one.
Also the Preprocess class could have a more descriptive name, like Compose (which is how torchvision names it).
The text was updated successfully, but these errors were encountered:
The preprocessing module is quite clunky and should allow a bit more freedom, such as
Transform
,is_observation_sequences
validation every time (as this is very costly).The use of
tqdm
progress bars is also questionable, and theverbose
argument clutters everything up.Even if progress bars are used, don't always make them full-width.
Finally, the interface for the
Transform
class could be cleaned up, particularly when it comes to having to define a nested function in thetransform
method.Instead of:
The user should just define the transform as an instance method operating on a single observation sequence:
then there is no need for
transform()
, and__call__()
would handle the call to_apply()
, and_apply()
would have access to_transform()
directly instead of needing to pass it as an argument.Note: Would need to think of a way to handle transformations that require all of the observation sequences rather than just one.
Also the
Preprocess
class could have a more descriptive name, likeCompose
(which is howtorchvision
names it).The text was updated successfully, but these errors were encountered: