Weisu Yin

demo page

5

Adds interactive demo through camera in the landing page

Currently, we use string concatenation and os.path.sep many places instead of doing os.path.join. This could be error prone.

Notebook server creation panel

1

The current notebook creation page is confusing because we are using the pod setting to configure the pod instead of the UI, but the UI are still there showing different...

While renaming the file/folder, you don’t see your keystrokes being reflected on the screen(but they are actually being typed in)

Tensorboard connection timeout

5

Tensorboard is not able to connect. We assume there is some port we need to open in order to make tensorboard to work.

Docker build out of space

3

Code Build will run into out of space issue if the image already exists because it will try to pull the image and build on cache.

Adding capability of taking auxiliary data

7

The idea of this PR is to enable benchmark yaml to take in auxiliary data. Auxiliary data could be useful when the task requires more than train and test datasets....

Keeping object dtype and string dtype

6

AMLB is converting object dtype and string dtype to category dtype: https://github.com/openml/automlbenchmark/blob/master/amlb/datasets/file.py#L310-L312 This would bring trouble when the framework requires a certain dtype to operate properly. For example, autogluon multimodality...

Loading test dataset failed with inferred data type from train dataset

5

Currently the logic when loading a dataset involving use pre-determined data types: [_ensure_loaded](https://github.com/openml/automlbenchmark/blob/master/amlb/datasets/file.py#L305-L320) This would cause loading to failure when the inferred data type is incorrect. In my case, I...