Closed Innixma closed 1 year ago
@sebhrusen @PGijsbers
Some questions I have:
[Solved in Innixma/automlbenchmark/pull/6] Is there a way to specify information such as prediction_length=5
on a per dataset basis? prediction_length is the look-ahead requirement for prediction and dictates the difficulty of the task. I'm wondering if I can specify it as part of the yaml file definition of the dataset in ts.yaml
. Ditto for a couple other things like timestamp_column="Date"
and item_id="name"
.
[Solved in Innixma/automlbenchmark/pull/6] How can I update and specify the logic that does the final scoring based on predictions and truth? It needs to be altered to work with TimeSeries. Additionally, it may take a different form, such as if the metric requires quantile predictions to calculate.
@sebhrusen I'd appreciate it if you can have a look, I have very limited availability due to a paper deadline.
@PGijsbers sunny holidays right now: will look at it when I'm back next week.
Update: Several of the TODO / FIXME comments have been addressed by @limpbot in https://github.com/Innixma/automlbenchmark/pull/6
Code example:
python3 runbenchmark.py autogluonts ts test
Log output:
@sebhrusen Sorry to ping but would you be interested in reviewing this PR? A large chunk of the logic was written by @limpbot who is interning with us currently, and it would be great if he received feedback so as not to block his time-series benchmarking efforts.
@Innixma I'm looking at it now and will make a full review before Monday. Outside implementation details/modularity, I mainly want to be sure that it is not designed to first satisfy AG's timeseries implementation and can be generalized to other implementations (would be nice to have an alternative implementation): for now to satisfy your needs, I'll mainly ensure that the changes are limited to data loading + AG implementation as much as possible.
Sounds good! I agree that we should make sure the input/ouput/scoring definitions are generic and not AG specific. perhaps the AutoPyTorch-TimeSeries folks (@dengdifan) would be interesting in reviewing / trying to add on their AutoML system as a framework extension to this logic?
Thanks @sebhrusen for the detailed review!
@limpbot would you like to have a go at addressing some of the comments? Feel free to send a PR to my branch as you did in your prior update.
I merged @limpbot's changes into this branch via his PR: https://github.com/Innixma/automlbenchmark/pull/7
@sebhrusen The branch should be ready for 2nd round of review.
Thanks @sebhrusen for the detailed review! @limpbot has addressed some final comments in the latest update, which should also fix the autogluon.tabular error you mentioned.
I think it will be interesting for us (cc: @PGijsbers) to start thinking about supporting new kind of tasks
I missed the "mention" ping (just thought it said a "subscribed"), sorry I didn't check earlier. Definitely, I want to first wait for the JMLR reviews and finish "that part of the project", but creating a more flexible environment for people to add new types of tasks would be a great next thing that invites more people to use (and extend) the benchmark tool.
Thanks Innixma and Limpbot for your contribution 🎉
[Don't merge this PR]
This PR is a proof of concept of time series data and framework support in AutoMLBenchmark.
To run, follow the instructions in the newly added
frameworks/AutoGluonTS/README.md
.