Benchmark DeepMatcher Data Sets

[abrinkmann]

Hey,

I would like to experiment with your framework and the deepmatcher benchmark data sets.
Unfortunately, the splits of these data sets are not along clusters as your framework expects it.
Is there any easy workaround to run your framework with the correct splits of the deepmatcher data sets?

[fjsj]

Hi @abrinkmann, the no-clusters branch has some experimentation on training based on pairs instead of clusters. But I don't recall if results were good enough. Please check it and LMK if you have any questions.

If you're experimenting this for publishing some paper, let me know too.

[abrinkmann]

Hi @fjsj, great thanks. I will have a look at the branch and experiment with the code :-)
I'll keep you posted.