Skip to content

LC0 abrupt change of mind (bug ?) #2190

@Ajosse7

Description

@Ajosse7

Hi
This issue comes from the analysis of a position
rnbqk2r/pp1p1ppp/4pn2/4P3/1b1N4/2N5/PPP2PPP/R1BQKB1R b KQkq - 0 6

Lc0 v0.31.2 considers Ne4 (-0.77) at depth 18.
One plays Ne4.
LC0 considers Qd3 (+0.71) at depth 24.
Ok logical.

White plays Qg4 and there the evaluation drops down nearly instantly at (-2.43).

I understand that good variations are difficult too find.
However I find curious that the program sticks with the bad move for so long.
It nearly never finds Qg4, but when forced to play Qg4 realizes nearly instantly that the move is good.

Strange ? or not ?
Monte-Carlo consequences ? All the more interesting as Stockfish finds Qg4 quickly, where LC0 misses.

Regards
Eric

Metadata

Metadata

Assignees

No one assigned

    Labels

    blunderBlunders and suboptimal play

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions