How many nodes/evals are too many?

[stevemulligan]

My engine is standard alpha beta search with quiescent search and transposition tables. I want to know how I compare to other engines in terms of total nodes searched and evals performed but I'm not sure which engine to compare against.

Eg, Let's say my search looks at 15640 nodes from the start position and Gaviota looks at 2714. I know it has Null move and I don't but that's about all I know about Gaviota.

A chart of approx node counts & evals expected in the type of search I'm using would really help me figure out where I need to direct my efforts.

[michiguel]

My engine is standard alpha beta search with quiescent search and transposition tables. I want to know how I compare to other engines in terms of total nodes searched and evals performed but I'm not sure which engine to compare against.

Eg, Let's say my search looks at 15640 nodes from the start position and Gaviota looks at 2714. I know it has Null move and I don't but that's about all I know about Gaviota.

A chart of approx node counts & evals expected in the type of search I'm using would really help me figure out where I need to direct my efforts.

Gaviota has also LMR, which reduces the number of nodes significantly. but also have extensions, which makes things difficult to compare.

I would compare the engine with a proven good, bug free, but simple engine. I think that Fruit 1.0 would be a good idea. You can switch off null move with a UCI parameter, so you can make a good comparison. I used to use it for that purpose.

Take into account that sometimes evaluation has a big impact on tree size, so comparisons are only "ball park".

Miguel

[stevemulligan]

Ok for Fruit 2.1 from startpos to depth 5,

all pruning off - 10398 nodes.
only null move pruning - 1934 nodes. Nice.


That gives me some nice targets to shoot for. Going to try null move first based on this page (from Glaurung Chess for iPhone website).

I'm not sure what the search code example is based on (stockfish maybe?) but this little bit of code explained null move for me:

Code: Select all

// Null move search:
  if(ok_to_do_nullmove_at_this_node()) {
    make_nullmove();
    value = -search(-beta, -beta, -(beta-1), depth-4);
    unmake_nullmove();
    if(value >= beta) return value;
  }

I had spent hours reading descriptions of null move and never really got it until I saw that snippet.

[michiguel]

Ok for Fruit 2.1 from startpos to depth 5,

all pruning off - 10398 nodes.
only null move pruning - 1934 nodes. Nice.


That gives me some nice targets to shoot for. Going to try null move first based on this page (from Glaurung Chess for iPhone website).

I'm not sure what the search code example is based on (stockfish maybe?) but this little bit of code explained null move for me:

Code: Select all

// Null move search:
  if(ok_to_do_nullmove_at_this_node()) {
    make_nullmove();
    value = -search(-beta, -beta, -(beta-1), depth-4);
    unmake_nullmove();
    if(value >= beta) return value;
  }

I had spent hours reading descriptions of null move and never really got it until I saw that snippet.

I was referring to Fruit 1.0 available here
http://wbec-ridderkerk.nl/html/download.htm

It is simpler, so it would be a more realistic target at the beginning (I think).

Miguel

[stevemulligan]

Sorry, grabbed the linux version out of habit.

Fruit 1.0 gives 16091 (with null move and aspiration search off - I think its very similar to what I'm using). That makes me feel a lot better.