Very nice article - I discovered the Lichess opening explorer only recently and this gives ideas on how to use it. I wonder or mildly disagree about one point: "you’re unlikely to have a large sample size in a given opening":
I play on Lichess since the start of the pandemic and now have about 2500 games (there are certainly more active players). These include 396 on the white side of the Sicilian and 618 with black. Of course it then splits up, but still 20-50 games after things "clarified", eg. 47 with white at the start of the Kalashnikov (after 4.-e5) or 33 on the black side of the Morra gambit.
So what would you consider a sufficiently large sample size? I ask the chess player and coach on what might be meaningful, not the data scientist about statistical significance.
Three years is a long time for teenagers (I was a youth coach myself) or young adults, or for players who are rapidly improving or declining, but not really for someone like me - middle-aged, stuck at or maintaining a certain chess level? I keep working on my openings and then do or don't remember what I have studied ... .
The tradeoff of having this many games: it's blitz, often decided by things unrelated to opening outcome or middlegame position - exchange of blunders, losing on time in winning positions, or flagging the opponent in a lost position. So one still has to check the actual games on HOW the result arose (a coach may help), and relevance of blitz for classical chess may be limited. And for me, my team avoiding relegation in a German league is more important than whatever happens on Lichess - thus I will do targeted prep for one opponent in the last round.
Very nice article - I discovered the Lichess opening explorer only recently and this gives ideas on how to use it. I wonder or mildly disagree about one point: "you’re unlikely to have a large sample size in a given opening":
I play on Lichess since the start of the pandemic and now have about 2500 games (there are certainly more active players). These include 396 on the white side of the Sicilian and 618 with black. Of course it then splits up, but still 20-50 games after things "clarified", eg. 47 with white at the start of the Kalashnikov (after 4.-e5) or 33 on the black side of the Morra gambit.
So what would you consider a sufficiently large sample size? I ask the chess player and coach on what might be meaningful, not the data scientist about statistical significance.
Three years is a long time for teenagers (I was a youth coach myself) or young adults, or for players who are rapidly improving or declining, but not really for someone like me - middle-aged, stuck at or maintaining a certain chess level? I keep working on my openings and then do or don't remember what I have studied ... .
The tradeoff of having this many games: it's blitz, often decided by things unrelated to opening outcome or middlegame position - exchange of blunders, losing on time in winning positions, or flagging the opponent in a lost position. So one still has to check the actual games on HOW the result arose (a coach may help), and relevance of blitz for classical chess may be limited. And for me, my team avoiding relegation in a German league is more important than whatever happens on Lichess - thus I will do targeted prep for one opponent in the last round.