Patterns across various versions of LeelaZero training

[Garf]

The last example in this quick overview is the pattern that is #1000 in global popularity: invasion at 1x1 against the opponent's 4x4! Interestingly, this still comes up in relatively recent versions, every now and then. That happens very rarely (about once in 250K patterns, and is mostly likely due to the artificial randomization for the fist 30 moves in self-play training games) but it is not "dead" yet in recent versions - unlike the pattern above - so it looks like LeelaZero still has some hopes for this variation.

This has been remarked several times, but maybe it's worth repeating here: the first line contains more go board squares than any other, so there's proportionally more chance of playing a first line move if a move is picked nearly randomly. That's why this is still appearing in training (but no match) games, and also why it has not died out, unlike some other patterns, whose "luck" odds just worse, i.e. it is the most-symmetries gathering first-line response to the most popular pattern.

Excellent work BTW. Can't wait to see the next update.

[sorin]

This has been remarked several times, but maybe it's worth repeating here: the first line contains more go board squares than any other, so there's proportionally more chance of playing a first line move if a move is picked nearly randomly. That's why this is still appearing in training (but no match) games, and also why it has not died out, unlike some other patterns, whose "luck" odds just worse, i.e. it is the most-symmetries gathering first-line response to the most popular pattern.
Excellent work BTW. Can't wait to see the next update.

Your explanation makes sense indeed, I didn't think about this argument before!

I re-did my analysis, but now I realize that I don't know exactly what to publish.
Basically I have the following data:

• "pattern view": top-1000 most popular patterns (across all versions), each with a popularity timeline graph
• "version view": version-by-version (in chronological order), top-10 (or so) patterns that are most popular in that given version and have not been top-10 in any older version (to prevent duplicates)

I am looking at this data but it just doesn't seem to be very interesting to publish as an update to the previous ones, it seems to be "too much data" and too little differences from previous data.
Any suggestions for what would be interesting to show, better than the two "tons of patterns and timelines" sets I described above?

[ez4u]

...

Excellent work BTW. Can't wait to see the next update.

I second both statements here!

[Garf]

For me the most interesting to see is how the bot likes (or dislikes) the normal human moves over time, and the same for the moves from the Alpha Go Zero games. Where has it started rediscovering theory, and where is it inventing new things...

[Uberdude]

sorin, someone has noticed that LZ seems to have gone off the early 3-3 invasions somewhat: https://www.reddit.com/r/cbaduk/comment ... _in_leela/.

Give it a week or two for random fluctuations to even out and then re-running your analysis on 3-3 invasion and related patterns would be interesting.

[sorin]

I just learned about this project that looks nicer than my old analysis: http://zero.sjeng.org/opening
Please note the different tabs for different start moves.

Also, somebody else-else did yet another very nice short analysis described here: https://www.reddit.com/r/cbaduk/comment ... rd_lately/

Like Uberdude wrote, LeelaZero is less crazy about early 3x3 invasions, and started playing keima kakari more.

[sorin]

I re-ran my analysis on LeelaZero versions up to May 21st or so, and wrote an article with focus on the versions before/after ELF came along, using some hand-picked pattern that I found interesting: http://361points.com/media/gopatterns// ... r_elf.html

Sorry I didn't get to address some of the earlier suggestion from Uberdude (to attach info about the level of various networks), since I realized I don't now how to do that; I can collect the LZ-Elo numbers, but those don't mean much, and to translate them into human kyu/dan seems hard).

Also, another suggestion I don't know how to address is the one from Garf, about finding when it is rediscovering existing theory and when it is innovating - I try to point out when some pattern is familiar, vs when it's "alien", but I feel that is kind of superficial, some more in-depth analysis would be needed.