-
Notifications
You must be signed in to change notification settings - Fork 211
Conversation
Hello @tchaton! Thanks for updating this PR. There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻 Comment last updated at 2021-03-29 18:33:32 UTC |
…tning-flash into datapipeline_poc_1
Codecov Report
@@ Coverage Diff @@
## master #141 +/- ##
==========================================
+ Coverage 76.52% 79.89% +3.37%
==========================================
Files 56 55 -1
Lines 2334 2447 +113
==========================================
+ Hits 1786 1955 +169
+ Misses 548 492 -56
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
@tchaton @justusschock can we get this done asap as it blocks transition to proper PL version |
@Borda why does it block? This should be independent. |
pls, check #133 (comment) |
What does this PR do?
This PR introduces the new API for DataPipeline.
Objective:
Provide a flexible API which organise user processing code toward higher readability, debugging and performance.
DataPipeline are composed from 2 parts: Preprocess and Postprocess.
Preprocess implements the following hooks:
Postprocess implements the following hooks:
The DataPipeline are aware of the Trainer RunningStage, meaning they know if they are running training, validation, testing, predicting,
The users can customise each hooks for a specific RunningStage by adding
train
,validation
,test
,predict
as prefix before every hooks: Example.train_load_data
function would be used for Training stage only or use booleanself.training
,self.validating
,self.testing
andself.predicting
TODOs:
per_sample_pre_tensor_transform
,per_sample_to_tensor_transform
,per_sample_post_tensor_transform
Before submitting
PR review
Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.
Did you have fun?
Make sure you had fun coding 🙃