How to reproduce filter perfermance ?

[chinaux]

https://github.com/zilliztech/VectorDBBench/issues

I have 3 questions, please help confirm：

1. Which dataset is used, cohere 768 1M/10M?
2. Has an inverted index been constructed for the fields' id 'and' label '?
3. What does the filtering ratio in the legend mean? For example, 98%, do those who meet the filtering criteria account for 98% of the total, or 2%?

[xiaofan-luan]

1. yes we use cohere but actually you can test with any dataset
2. we are not building any index, purely by default setup
3. 98% means only 2% data is left

[chinaux]

Thanks for the reply