Gin

Thank you @kts for opening the issue & Sorry for the late reply! re: your issue and blog: https://kenschutte.com/gzip-knn-paper/ When the flag `rand=False`, the logic of the code _is_ to...

Hi @YannDubs, wow thanks for pointing this out!!! I was only aware of the dataset issue of DengueFilipino. Thanks @kts for verifying the huggingface dataset issue. People should be aware...

Here are results using the original DengueFilipino dataset. I also checked the original Kirundi dataset, it still has the data contamination issue.

Hi @kts, thanks for your reply. No worries, I think it's my responsibility to clarify things : ) I'm glad we are on the same page about the top k...

Hi @flipz357, Thanks for your comments. I agree with you that different methods will have different likelihood on tie-situation. So yes, I will report the accuracy with random selection.

hey @kts @eric-xw I understand that you think using the max accuracy of knn for comparison is unfair. What I was trying to say is that random guess has its...

> Why report the **max** rather than the **mean**? Because reporting **mean** induces more stochasticity and/or magnitude more of experiments. As you see, we have few-shot experiments across all the...

thanks @cyrilou242, what you said makes sense. I will also include minimum accuracy and k=1 case.

@cyrilou242 here are results with both lowerboud and upperbound. @kts I also include the random (only run once though) for reference. @flipz357 As you mentioned it may be unfair to...

hey @maxmax1992, sure I do agree it makes the accuracy of gzip inflated. So please check the updated version of the table  The table also includes the updated result...