New workload type: DATAGEN
AndyGrant opened this issue · 1 comments
AndyGrant commented
The GenFens interface is finished in Torch. The result is that public engines can be executed like follows, to generate openings: ./Ethereal "genfens N seed S book <?book|None>"
. Private engines can be run via ./Torch "setoption name EvalFile value ..." "genfens N seed S book <?book|None"
. In both cases, the OpenBench client will expect N lines of output starting with "info string genfens ".
The following steps will finalize this process:
- Add a new workload type to the workload creation page called
DATAGEN
. - In most cases, treat these exactly the same as
GAMES
workloads. Consider usingEnum
- However, use the trinomial modeling and not the pentanomial modeling
- In the client, remove the -repeat argument to Cutechess
- Add a utility to execute many copies of the engine at once, to run genfens, and collect them.
AndyGrant commented
Resolved a while ago, but no existing demonstration in a public engine is present.