Wonderful analysis, I believe Alireza did something similar but thanks for producing the statistics
Label your freakin axes man. Twice I had to go look up some other article you linked to try to figure out what an axis represented. I shouldn't have to do that. A graph should be self-explanatory, or there should be a short summary in the text.
The "number of book moves" graph y-axis should be labeled "average number of moves played per game." You also need a comment about the colored bars explaining that they represent how common a move is in your database. And in fact colored bars are a terrible way to represent that. It is numeric data that should just be plotted on a logarithmic scale (with its own axis label). Also it's not clear, either in this article or the linked one, how you would count a game that starts like "1. e4 c5 2. Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3" ? These are all book moves, does this game count *once* in your chart or does it count *once per each position* (so 5 times)? I'd assume it's the second one but you have to make these things clear.
The accuracy distribution axes should also be labeled. I think the appropriate x label is "average centipawn loss per game," and the appropriate y label is "fraction of games." But frankly I don't know, maybe you are averaging over moves or something. Label your axes.
The "number of book moves" graph y-axis should be labeled "average number of moves played per game." You also need a comment about the colored bars explaining that they represent how common a move is in your database. And in fact colored bars are a terrible way to represent that. It is numeric data that should just be plotted on a logarithmic scale (with its own axis label). Also it's not clear, either in this article or the linked one, how you would count a game that starts like "1. e4 c5 2. Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3" ? These are all book moves, does this game count *once* in your chart or does it count *once per each position* (so 5 times)? I'd assume it's the second one but you have to make these things clear.
The accuracy distribution axes should also be labeled. I think the appropriate x label is "average centipawn loss per game," and the appropriate y label is "fraction of games." But frankly I don't know, maybe you are averaging over moves or something. Label your axes.
@Berder
I'll certainly label my graphs better in the future.
> You also need a comment about the colored bars explaining that they represent how common a move is in your database. And in fact colored bars are a terrible way to represent that. It is numeric data that should just be plotted on a logarithmic scale (with its own axis label).
The legend of the graph says what the colour of the bars means and I decided against always repeating what the graphs mean if I've written about them previously. But maybe I should be more clear about that.
I'm unsure what you mean by "It is numeric data that should just be plotted on a logarithmic scale".
> Also it's not clear, either in this article or the linked one, how you would count a game that starts like "1. e4 c5 2. Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3" ? These are all book moves, does this game count *once* in your chart or does it count *once per each position* (so 5 times)? I'd assume it's the second one but you have to make these things clear.
The plot shows the average number of moves with more than 10,000, 1,000 etc games in the database. So in your example, if the position after 5 moves still appears more than 10,000 times, then the 5 moves count towards the average number of moves with more than 10,000 games (and hence also for more than 1,000, 100, 10 and 1 games).
I'll certainly label my graphs better in the future.
> You also need a comment about the colored bars explaining that they represent how common a move is in your database. And in fact colored bars are a terrible way to represent that. It is numeric data that should just be plotted on a logarithmic scale (with its own axis label).
The legend of the graph says what the colour of the bars means and I decided against always repeating what the graphs mean if I've written about them previously. But maybe I should be more clear about that.
I'm unsure what you mean by "It is numeric data that should just be plotted on a logarithmic scale".
> Also it's not clear, either in this article or the linked one, how you would count a game that starts like "1. e4 c5 2. Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3" ? These are all book moves, does this game count *once* in your chart or does it count *once per each position* (so 5 times)? I'd assume it's the second one but you have to make these things clear.
The plot shows the average number of moves with more than 10,000, 1,000 etc games in the database. So in your example, if the position after 5 moves still appears more than 10,000 times, then the 5 moves count towards the average number of moves with more than 10,000 games (and hence also for more than 1,000, 100, 10 and 1 games).
true