Results for 1M+ labels

[shawank]

Thanks for releasing the code. I have been able to successfully reproduce results for standard datasets such as EURLex-4K, Amazon-670K. Unfortunately, I am getting nearly random predictions for dataset above million labels (Amazon-3M to be specific). Do I need to use some different set of parameters like number of buckets etc.?

[Tharun24]

Hi Ankush, Thanks for working on my code. Above a million labels, we need to increase the number of buckets B and hidden layer size. Number of repetitions R can still be around 16 or 32. Changing B will require you to rebuild the hashtables. Regards, Tharun Medini

…

Sent from my iPhone

On Feb 5, 2020, at 11:30 AM, Ankush Shaw ***@***.***> wrote:  Thanks for releasing the code. I have been able to successfully reproduce results for standard datasets such as EURLex-4K, Amazon-670K. Unfortunately, I am getting nearly random predictions for dataset above million labels. Do I need to use some different set of parameters like number of buckets etc.? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.