| |
BGonline.org Forums
Best online data source for studying opening replies?
Posted By: Timothy Chow In Response To: Best online data source for studying opening replies? (Matt Ryder)
Date: Wednesday, 23 December 2009, at 3:16 p.m.
Having recently gone through the process of studying opening replies in some detail, I have some thoughts to share.
1. First, if you don't already know about it, you may find my one-page summary helpful. It is based on Stick's rollouts.
2. As a general rule, cubeful rollouts are to be preferred to cubeless rollouts, unless you have some reason to suspect that the bot is not handling the cube correctly. GNU 2-ply cube is certainly not perfect, as Michael Depreli's ongoing bot comparison project is demonstrating, but what matters most are cube errors one or two turns after the position you are rolling out, and for money-game opening replies, we're probably O.K. because GNU 2-ply is probably making few cube errors on the 3rd or 4th roll.
3. As for the quandary of which bot to choose, that is just a fact of life that you have to deal with. We don't know the absolute truth. On your first pass through, when you are just trying to memorize some information in order to get a foothold on this material, I would recommend making an arbitrary choice. For each opening reply where there is a difference of opinion, just memorize one acceptable play. The purpose of this first pass of memorization is to familiarize yourself with surprising plays (like how to play 3-2 after your opponent opens with 13/8 6/5) and to force yourself to look for patterns in the data. Memorizing 400 random facts is much harder than memorizing 50 patterns, so in the course of trying to memorize things you'll inevitably find yourself seeking out patterns.
Here's an example. Say you're trying to learn to play 2-1 on the second roll. Do you blindly memorize 30 separate random facts? No. You'll remember something like this: "I'll hit if I can. If I still have a die to play, I'll use it to bring a checker down from the midpoint, unless that's impossible, in which case I'll use it to split. If I can't hit, then I'll slot, unless that would expose me to a double shot or unless my opponent is already priming me by making the 5-point, 4-point, or 3-point; then I'll split with 13/11 24/23." There will be a couple of exceptions to this set of rules, of course. If your opponent opens with 24/16 then 24/21 is a better reply than 13/11 24/23. And as has been discussed recently on this forum, slotting may be the correct reply to 13/8 24/23. But here's the point: Your ultimate goal is not to memorize 400 separate facts anyway. Your goal is to develop an understanding of the opening, with bot data as a means towards that end. It's more important that you start observing these patterns and making sense of them than that your plays match some bot's play in every last detail. Towards that end, making an arbitrary choice in the few cases where there's still no consensus is not going to hurt you.
On your second pass through the data, you can then start to worry about the exceptional cases. Why are these exceptions? Is there a subtler issue that I'm missing, or is it a bot weakness? Again, varying opinions by different bots should not be viewed as an annoying impediment to perfect memorization, but as an opportunity to explore some aspect of opening theory that might not otherwise have been apparent to you.
The bottom line is, you should free yourself from the error of thinking of the bot data as a monolithic corpus to be memorized and parroted back. Memorization is an important exercise, but it is only a means towards the end of deeper understanding.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.