Training problems

[janlehky]

When I try to start training on IRIS dataset I always get code 500 from API.

It tries to do text clasification on numerical data.

[fatkaratekid]

By default the system is set to work only with NLP problems. Given the limited resources so far, it was better to have one thing done right than to make it work a little bit for everything. The import assumes that your data is textual.

In order to make this work, I would suggest taking on this task #10. This is a prerequisite for what you want.

Possible approach:

1. Look at the DataFrame's column type, after the file has been imported, and store that information in the Datasets class in boonai/boonai/model.py.
2. Define an algorithm that accepts integers as inputs in here boonai/boonai/project/api/machine_learning/algorithm_selection.py. So, add algorithm, add description (a new field needs to be added there, and maybe even two more, one for tags and one for category).
3. The trained model needs to store which input fields it accepts. So Models in boonai/boonai/model.py has to be updated (again, maybe adding category and tags fields).
4. Update site to show only compatible algorithms when dataset is selected for training.
5. Update site to show only compatible model when data is selected (and/or vice versa). Maybe a reset button needs to be added, so you don't end up in a position where you cannot select anything new anymore.

I assume you are using the SQLite, so don't forget to delete it after you have made changes. SQLAlchemy will automatically create the new database.

[janlehky]

Understood.

I will see if I can help with issue #10

[fatkaratekid]

That would be a great contribution!