Lichess says the best engine move h6 is bad

@KMcGeoch said ^

So then you have a trade off where is it worth spending 4x CPU resources for an issue that could affect 0.1% of suggestions. Considering the costs of cloud compute the marginal benefits aren't worth the slight accuracy gains. I'll also add that if Lichess didn't expand infrastructure that game analysis would also get slower.

im confused , usually the feature calculates a move for 200ms on my cpu but in this case it said instantly bad move , so did it take that information from the server game analysis ? if yes why only sometimes ?

I think that's difference between it being processed locally and server side. If you turn on the engine it calculates it on your computer while full game analysis is done server side. Spending 200ms on a move on a local device isn't a massive load. Spending that time on every game played on the server is.

@KMcGeoch said ^

I think that's difference between it being processed locally and server side. If you turn on the engine it calculates it on your computer while full game analysis is done server side. Spending 200ms on a move on a local device isn't a massive load. Spending that time on every game played on the server is.

i understand that server analysis is bad in this case , but why did it here use cloud analysis and usually user cpu analysis ?

I think the rationale is that if both players are going to go over the game afterwards then cloud analysis makes sense and both players can view the results. If it was done on the basis of user cpu analysis then you would have issue that both users could obtain different results depending on the hardware they're using. It would also duplicate amount of processing and if a player closes the tab or browser you'd have to handle error handling there.

However when you analyse a separate variation you're curious about then you're only person interested in the result so it makes sense for it to be local.

At least when I review games the computer analysis is mainly there to pick up obvious errors and is a quick initial sort. If I want to look in more depth then I'll use user cpu analysis and see lines in more depth.

@KMcGeoch said ^

I think the rationale is that if both players are going to go over the game afterwards then cloud analysis makes sense and both players can view the results. If it was done on the basis of user cpu analysis then you would have issue that both users could obtain different results depending on the hardware they're using. It would also duplicate amount of processing and if a player closes the tab or ...

ok , makes sense but why does the feature often need to cpu analyze a move if its supposed to know by server analysis ( if i understood it correctly )

Server analysis just analyses the moves played in the game and when it believes a player has made a blunder/mistake it then gives an additional line it believes is best although depth isn't huge.

What it doesn't do is save analysis for every candidate move nor does it show the branches from the main line. If there are 3 equivalent lines only one is covered and if you wish to see why an idea used on the game/alternative line works/doesn't work in the other line you need to check moves and do user cpu analysis.

thanks for the explanations

analyzing by the cpu for only 200 ms is questionable but i guess its not better with cloud analysis