Why use -cost get the record of bigger latency?

[yogurfrul]

We know the cost in auto_schedule measure result means [latency] or [Cost]， the smaller the better.
While in the code as below:

lorien/lorien/dialect/tvm_dial/auto_scheduler_dial/result.py

Line 81 in bcd3913

# Push with -cost as heapq is min-heap as we want the worst record on the top.

as you say eapq is min-heap the record on the top would be the best.
Why use -cost get the record of bigger latency?

[yogurfrul]

@comaniac

[comaniac]

The heap aims to keep the top-N records with minimum cost. To do so, we iterate all records and push them to the heap. When the heap size is larger than N, we remove the one with worst cost. Thus, we need max-heap to keep the so far worst record on the top in O(1).

[yogurfrul]

Yes, I got the use of heap .
While I found the func records.peak() in tutorials to get the best schedule, actually it retuen the first record.
As you say above, the heap aims to keep the worst record on the top.
And I run the test with my own record of the same workload, the peak func really get the the first one in _data with worse cost.