Features Request list

[AakashKumarNain]

I tried out waveml last week and since I am actively developing an app, I have figure out certain features that are needed from an end user/developer perspective. This list isn't exhaustive but it's a good starting point:

1. The pypi package doesn't install all the dependencies. This should be mentioned in the Installation section and the links to download other wheel files should also be provided.
2. A better way to set all the required environment variables, especially for the standalone waveml test
3. Examples to showcase how to wrap model training using waveml in a wave app. This should include the Do's and Don't s
4. Training: When a user starts training a model using waveml, the API should provide some amount of logging, for example: training status (including ETA), DAI messages about datasets like leakage and all, etc. This helps in providing an interactive feedback to the end user, so that the user knows that the training is happening
5. Prediction: Provide examples for saving the endpoint_url. Although this seems trivial, it becomes important to keep track of all the trained models in order to compare their performance on a particular dataset
6. Datasets: Ability to use datasets from a s3 bucket

[vopani]

1. It's mentioned in https://github.com/h2oai/wave-ml/wiki/Quickstart but I agree it needs to be easier to find. Maybe we should add basic installation guide in main README? @geomodular
2. Should be part of Allow DAI / Steam details to be specified in runtime #39
3. Will cover in Improve Wave ML examples #43
4. It already does. You can see training progress for H2O-3 as well as DAI. Or are you asking for something else / more? If so, it should be done at H2O-3 or DAI client level and it will automatically reflect in WaveML.
5. Will cover in Improve Wave ML examples #43
6. Tracked Support dataset inputs from S3 (and other sources) #48

[AakashKumarNain]

Thanks @vopani for the update

It already does Yes, I saw that but that's stdout. Is there a way to capture that log information in a variable/file? I am asking this because a wave user won't see that info in the app until unless it is captured somewhere and then displayed in the UI

[vopani]

Ahh ok, I see how this can be useful. So, the DAI Client does support this using async create, for example: http://docs.h2o.ai/driverless-ai/pyclient/docs/html/concepts.html#example but WaveML doesn't use the async version.

This needs to be thought through. Maybe can track as a separate discussion.

[geomodular]

It's mentioned in https://github.com/h2oai/wave-ml/wiki/Quickstart but I agree it needs to be easier to find. Maybe we should add basic installation guide in main README? @geomodular

Yes, will create an installation section in README.

[geomodular]

The readme file is updated: https://github.com/h2oai/wave-ml#installation