Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Problems when adding analysis to pre-analyzed SGF #820

Closed
hope366 opened this issue Nov 26, 2020 · 2 comments · May be fixed by #821
Closed

Problems when adding analysis to pre-analyzed SGF #820

hope366 opened this issue Nov 26, 2020 · 2 comments · May be fixed by #821

Comments

@hope366
Copy link

hope366 commented Nov 26, 2020

I analyzed the SGF file using lizzie.
We analyzed from the first move to the last move. The average amount of analysis is around 500 playouts. Since it is about 500 visits / s, it takes about 1 second for each phase. I saved this.
Next, I wanted to add the amount of analysis and save it by overwriting.
Load the saved SGF and add the analysis.
At this time, I noticed something strange.
In one situation, the value on the board changes about 1 second after the analysis is turned on, but in another situation, it takes more than 10 seconds until the value on the board changes.

@hope366
Copy link
Author

hope366 commented Nov 26, 2020

It seems that the information displayed on the board regarding the playouts numbers may differ from the information displayed on the comment panel. Some aspects are in agreement and some are different.
If they are different, the numbers displayed in the comment panel will appear to be exactly 10 times the numbers displayed on the board (the total number of playouts displayed for all candidates).
Paste the screenshot as an example.
無題
The total number of candidates on the board is about 8.3k, but the number displayed on the comment panel is about 10 times that of 84k. The amount actually analyzed is about 8.3k displayed on the board. 84k is wrong.
And if you try to add the analysis amount in this situation, the numerical value on the board will not change unless you do the analysis amount of 84k. It looks like you are following the wrong information.
My PC spec is about 500 visits / s, so if it is 8.3k, I expect that the analysis will be added (the numerical value on the board will change) if I wait about 17 seconds, but in reality it is an analysis for 84k. The value does not change unless the amount (about 168 seconds) is applied.

@hope366
Copy link
Author

hope366 commented Nov 27, 2020

The strange phenomenon does not occur at the first overwrite, but seems to occur after the second time.
Also, even if the numbers on the board and the numbers on the comment panel are the same, it seems that the phenomenon of waiting for a considerable amount of time will occur.

http://www.kihuu.net/sgf/k00000130658.sgf
I used this to validate and found another issue.
Load this game record with lizzie and save it again soon. Then, the data at the beginning of SGF is rewritten.he date has changed and the player's name displayed below the board is no longer displayed properly (Chinese name)
The first data is ↓

(;EV[2020世界人工智能圍棋大賽預賽第1輪]DT[2020-09-27]PB[LeelaZero]PW[星陣圍棋]KM[6.5]RE[B+R]
無題1

The one saved after loading with lizzie is ↓

(;DT[2020-11-27]EV[2020世界人工智能圍棋大賽預賽第1輪]KM[6.5]PB[LeelaZero]RE[B+R]PW[星陣圍棋]SZ[19]AP[Lizzie: 0.7.4]
無題

kaorahi pushed a commit to kaorahi/lizzie that referenced this issue Nov 28, 2020
kaorahi pushed a commit to kaorahi/lizzie that referenced this issue Nov 29, 2020
toomasr added a commit to toomasr/lizzie that referenced this issue Dec 19, 2020
Fix featurecat#820 (playouts "1.5k" is wrongly read as 15000 in SGF)
@hope366 hope366 closed this as completed Dec 25, 2020
xiaoyifang pushed a commit to xiaoyifang/lizzie that referenced this issue Jan 22, 2021
xiaoyifang pushed a commit to xiaoyifang/lizzie that referenced this issue Jan 22, 2021
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging a pull request may close this issue.

1 participant