Validating on independent data


#1

I have trained a model on a dataset that I uploaded to the DLS. Now, I have a second dataset from an independet source and wish to run my model on that new data (hundreds of cases of known classes). The aim is to calc TP, FP, TN, FN and derived measures.

The data is in .npz format, therefore decribed in the csv file as two colums “filename” and “class”. Ideally, I would just like to use the data from within the platform, but this seems not to be possible (from the project tab, I cannot select another dataset, only a subset like test, or validation). So I deployed it, but then again, I have to give single file names in the API or web form. And I don’t even know if it will accept the name of a npz file instead of a png.

What are my options here?
Best, Markus


#2

Ah, I think I found it.

In the options, there is Upload. I’ll check this out. Apparently, this will re-upload a dataset as a test set, although it actually already exists as a training set – would be cool if I could just select it from the list of own datasets instead of re-uploading it…

Best, M


#3

would be cool if I could just select it from the list of own datasets instead of re-uploading it…

I would approve of this feature.


#4

You mean that in inference tab one should be able to select from already uploaded datasets?


#5

yes, exactly. i think it would be quite useful at times. For me at least it would.