LiuNeng
LiuNeng
@alexey-milovidov @nikitamikhaylov @kssenii
code ```c++ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include namespace DB { class FasterCompressedReadBufferBase { protected: ReadBuffer * compressed_in;...
data: https://1drv.ms/u/s!AjK3WtSMQ0kokhDc_1ThSIoUa_3H?e=l57NM6
@kitaisreal You can use the code above to reproduce, by modifying the code of the decompression algorithm for comparison.
The above statement does not explain my variable control over the decompression algorithm code, I made sure that only the decompression code was different between the control groups, and the...
> (in order of 1..5%; higher percentiles should be ignored because it was tested on laptop with running Slack and YouTube) The SQL I tested only had simple aggregations, and...
I don't have a solution to this problem, and I don't have any other ideas, the problem doesn't seem to have much impact in other types of queries
My test is corrected 😹😂
h_tv and w_tv are quadratic sum, their derivative is 1/2, in the backward, 2 * 1/2 =1 , many loss function use this tip.
应该是个中国人,导数1/2乘以2能抵消掉这样能减少小数计算.