World Shogi Forum

World Shogi Forum > Shogi in General > Analysing a game with engine

Analysing a game with engine
adelha (1434) 2017-04-21 04:05
Hello everyone 

I recently began analysing some of my shogi games with Gikou and there's something i don't understand about the evaluation of the position. it may sound silly.

What's the "score" ? i don't get why it gets negative or positive :(

http://system.81dojo.com/en/kifus/2207804

In this game for example at move 68, the score goes from -467 to -1071, we both activate our pieces so the position should be fairly equal ? 

I believe when the number is positive it means black is ahead and when it's negative it means white is ahead, but at move 43, the score is negative but it looks clear that my opponent has a great advantage. :/ 

Thank you for your time
Please login to post a comment.
3: kmullin (1065) 2017-04-22 01:35
For joseki stuff, there are Hidetchi's YouTube videos. But, he only covered the Quick Ishida variations and not the (slow) Ishida variations or the non-Ishida variations.

For the other variations, there are a few things.

(1)
Hosking (1997) _The Art Of Shogi_ has a good chapter,

and 

(2)
the old _Shogi_ magazine has a few articles on the topic.

I haven't checked carefully, but it might be that everything in the magazine is covered by Hosking. 

(3)
There's also the Nekomado _Joseki At A Glance_ book by Kitao. It has some info on non-Ishida closed-bishop-diagonal (Koyan, vs Super Rapid Attack) variations. However, it isn't very in depth. It is kind of like a series of flashcards with answers on the back. Still informative, but it is not as useful as Hosking.

Besides that you can go to professional games. Unfortunately, these are harder to understand without commentary on the moves that they play. But, it may still be useful. 'Yamajunn' has many games notated with opening and castle tags. Like this: 

http://yamajunnshoginews.blogspot.com/search/label/Sanken%20Bisha 

You just have to skip through the Ai Furibisha (Double Ranging) and Ishida ones to find Static Rook vs Third File Rook.

After that, you have to go to Japanese language material...
2: adelha (1434) 2017-04-21 05:40
Thank you so much kmullin for your answer, at the end i had an advantage of -3000 and i was very frightened because my king was exposed. evaluating position in shogi is very tough.

I'm very sad that there are not many materials for studying shogi like in go, i wanted to look for the joseki of sangenbisha x static rook to see if a silver crown castle was effective in this positon and how to activate my pieces better but i couldn't find anything, i guess i'll play it and play it and play it until i learn by myself. 
or maybe learn japanese and get a tesuji book :) 
Anyway, thank you kmullin, now i'll be more effective in my reviews, have a nice day :)
1: kmullin (1065) 2017-04-21 04:48 (Edited at 2017-04-21 04:50)
Scores > 0 favor Black, scores < 0 favor White.

The score is just a number generated by some math equation. We won't know exactly the equation unless we go through the engine's computer code. I think you can treat it as basically a black box that you can't see inside, it just generates a number based on what's inside the box. Whether or not the score corresponds to what you or anyone else thinks is 'favoring Black/White' is a research question really. Presumably, it *does* correspond somewhat to what humans think is good or bad for a given player in a given game. We don't know what the numbers mean really. It's just that generally a more positive number is more likely to better for Black and vice versa. We don't if a difference of 200 points is meaningfully different from 150 points or if 200 or –200 is meaningfully different from 0. But, a difference of 1000 points probably is significant and 1000 or –1000 is probably significantly different from 0. You can treat the numbers with 'a grain of salt' and pay more attention to overall trends and very large jumps/differences as being informative.

I see something slightly different for that game. I see a change from a –480 score for move #66 to –1155 for move #67 – using a 15 sec duration for evaluation. (Not between move #67 and #68 as you wrote.) Does this help explain the score? I think that Gikou finds that activating both players' rooks is not optimal for Black. Gikou suggests instead of ▲24銀 a move of ▲23歩打つ. So, the jump to a more negative score is probably explained by Black allowing White to take Black's pawn on 34 (as you did). Maybe it would have been better if Black did not allow that.

This is 81Dojo's web system for account and data management. The entrance to the shogi app can be found in the main site at http://81dojo.com.