Practical experiences & questions

[pat-s]

Hi guys,

I'll share with you my practical experiences from trying to use openml in the last days. Maybe it helps to improve the package :)

1. I started by uploading a CSV dataset as a dataset using the web interface at https://www.openml.org/new/data. No matter what I put in and how many fields I fileld out I always faced the error below. It worked using uploadOMLDataSet().

"Error 131: Problem validating uploaded description file - XML does not correspond to XSD schema. Please fill in all required (red) fields, upload a file or give a URL (not both), and avoid spaces in the dataset name."

2. I then tried the CL function uploadOMLDataSet(). Two points here: I was expecting to upload a dataset. A dataset for me is a CSV like file. However, the function actually expects what is a "task" object in mlr. If one is used to work with mlr, it is not logical to upload a mlr task as a "dataset" to openml. But maybe openML makes more sense here and the mlr logic lacks.

3. Creating an openML task using makeOMLTask could really need a better help page. How should the list for args parameters and input look like?

4. Once I created a task using makeOMLTask, how can I upload it? There is an upload function for datasets, flows and runs but none for tasks?

5. Is it possible to use a different resamp strategies in the web interface then the preselected ones? E.g. I would like to use "repeated spatial CV" from mlr or just "5-fold CV".

6. How can I "activate" my dataset? E.g. https://www.openml.org/d/41110