Backgammon Snowie

Snowie

Snowie vs. Jellyfish

From:   Kenneth M. Arnold
Address:   kenarnold@mindspring.com
Date:   30 May 1998
Subject:   Re: Snowie vs JellyFish
Forum:   rec.games.backgammon
Google:   3570792d.83246818@news.mindspring.com

My opinion: Snowie is light years ahead of JF in its interface and overall value. Snowie **probably** outplays JF, but that is not certain. Snowie surely plays backgames far , far better than JF. The real benefit of Snowie is that the interface is superb and makes it simple to improve one's game. Having used JF for years, I feel that Snowie taught me more in 3 weeks than JF did in 3 years. It highlights errors, blunders and cube decisions. It is possible to import entire matches from GamesGrid or Fibs or JF and then to have snowie roll out every single move in the match any number of times with just 2 or 3 mouse clicks. This means that you are compariong your plays to the 3888 game rollout instead of the Level 7 opinion which makes a huge difference in the quality of the information. Another interesting feature is that Snowie will provide (graphically and numerically) the skill level and 'luck level' for each player by game and by match. You can even build compilations of the skill level of each of your opponents over a period of time, which will tell you whether you are really getting outrolled or outplayed. In my opinion, anyone who wants to win at this game should get Snowie asap. The price is steep but it is reasonable to expect a major improvement in your game! You can view samples of the analysis the Snowie provides by going to the following site: http://wwwtech.cyberarts.com/web/snowie/matches/kavsjolev7pt.htm CyberArts is a distributor of Snowie.

David Montgomery writes:

I have tremendous regard for Snowie, but this post seems like quite a bit of hype. Snowie's interface is superior to JF's, overall, for the student of the game. It's not clear to me that it is for the average player. Snowie has a number of great features which do advance the state of the art for backgammon programs, but as of the most recent beta it still lacks some important features that JF has -- features important to the serious student of the game. I'm not sure I'm allowed to comment specifically on these, but in any event it wouldn't be fair to the developers of Snowie since it hasn't been officially released yet, so they can still add functionality. It is hype to say it's light years ahead. Maybe 1 year ahead. Two at the outside. (I know, light year is distance, years are time.) Snowie definitely plays some positions better than JF. JF definitely plays some positions better than Snowie. I know this for a fact, based on rollout data. Most of the time, the Snowie evaluations "look" a bit more accurate to me than the JF evaluations, but this is very subjective. I have seen positions where Snowie's evaluations were much more accurate than JF's, and positions where JF's were much more accurate than Snowie's. The best objective evidence I know of for comparing JF and SW's overall strength is the two program's average rating on FIBS. JF has several thousand experience points supporting the case that on level 7 with a modest time factor, JF's rating will average about 2050. Despite SW's 2100 rating, it passed 2050 less than 500 experience points ago. So based on the objective evidence, I think the claim should be, Snowie *probably* plays about as well, overall, as JF, but it might not. I'm not saying this is the best evidence we could have, just the best that I know of that has been gathered already. As far as a specific class of positions, backgames... well, backgames are actually of a multitude of forms. Based on the rollouts I have done, I feel that JF plays backgames better. JF definitely plays some of them better. Maybe I haven't looked at enough positions, or Ken is considering a different type of backgame class than I am, but I would be stunned to find out that SW plays backgames, overall, "far far better than JF" since I know that in the cases I've checked, JF plays them better. > The real benefit of Snowie is that the interface is superb and makes it > simple to improve one's game. Having used JF for years, I feel that > Snowie taught me more in 3 weeks than JF did in 3 years. This is hype. Snowie has a much, much better interface for reviewing matches and games, but the time ratio is nothing like 52:1. I have been reviewing matches with almost nothing but Snowie since I began beta testing it, but I've certainly gotten more benefit out of JF the last 4 years than out of snowie the last 2 months. I think the time advantage of Snowie is probably between 2:1 and 10:1 -- still big. > In my opinion, anyone who wants to win at this game should get Snowie > asap. The price is steep but it is reasonable to expect a major > improvement in your game! I agree that Snowie is going to be a great product, and that any backgammon player would like to have it. But I have heard rumors of an extraordinarly high price. On the other hand, many players feel that the only way for Snowie to get decent sales is to offer it at a price below JF. A number of players are looking forward to declining prices for top backgammon software now that there is competition. I guess time will tell. Based on the latest beta, I would recommend Snowie over JF, if price were no object. But I have gotten a lot of players started with computer backgammon, setting them on FIBS (free), teaching them to use svempa's great FIBS to JF .mat converter (free), and making sure they buy JF tutor ($110, if i remember right). For most players, this gives them more feedback than they will every absorb. If Snowie costs several hundred dollars, I don't see how I could recommend it over this setup for anyone to whom $100 isn't chump change. For the truly serious players, Snowie's price might be nothing more than their equity for one night in the local chouette. But I don't know how many of us there are. David Montgomery monty@cs.umd.edu monty on FIBS

Ken Arnold writes:

Hmmmm.... should have put this in the previous post. The 'opinion' levels in all of these programs are too weak to be of much value to an expert player. Only the rollouts are really reliable. Every time that I have played a long match against JF and then reviewed it, JF Levels 5 and 6 point out many errors which are not errors at all. Upon doing a Level 7 evaluation or a rollout JF reverses itself and selects my original play more often than not. Thus, JF has been of very little value to me. Perhaps you are more patient and more willing to do the work needed to get the most out of JF. I like to play a match against a top human player online, import it to Snowie, and then have Snowie do a 3888 rollout of every play in the match (which takes ovenight on a P-Pro 200) while I sleep. This method yields excellent data for just a minute or two of work. It becomes easy to review the entire match or just to focus on the rolls where there was an error. How long would it take to do the same in JF? More than 10 times longer (perhaps 50) and more time than I have the patience for. I have written, reviewed and tested a LOT of software over the last 17 years. This is an extraordinarily high quality release 1.0, better than any other software product that I have ever seen and I doubt that anyone can point out counterexamples.

David Montgomery writes:

> The 'opinion' levels in all of these programs are too weak to be of much > value to an expert player. Only the rollouts are really reliable. > Every time that I have played a long match against JF and then reviewed > it, JF Levels 5 and 6 point out many errors which are not errors at all. > Upon doing a Level 7 evaluation or a rollout JF reverses itself and > selects my original play more often than not. Thus, JF has been of very > little value to me. Well... I disagree with your first sentence, although I agree with most of the rest. First though, there isn't any reason to consider level 5 and 6 'opinion' levels and level 7 not. Level 7 opinions are just a double- click away. It is certainly true that the stronger the player, the less value in the evaluations. I have extensive experience with FIBS ratings, so I'll base my comments on that. First, note that the vast majority of players are not experts -- they are not even open level tournament players. For anyone who can't maintain a rating of 1750 on FIBS, I think the value of rollouts over evaluations is negligible (and certainly not worth the cost of the analyzer or SW, nor the hassle of setting up rollouts). If a player can't play at this level, they should read the good books and use JF or SW as a tutor until they improve, and the evaluations provide tremendous value to those capable of learning from them. As you move over 1800, evaluations will be seen to be wrong more and more often, relative to your own choices. However, even the best human players still occasionally make clear mistakes, even 'blunders' (as defined by default by SW: a .08 cubeless mistake). When SW or JF points out that you have made a play .05 worse, on their highest level of evaluation, you will reliably find that a rollout will confirm the result. There are exceptions, some occuring in easy to anticipate classes of positions, but even so, I would guess that a .05 error will show up as an error in a rollout 90% of the time. I have reliable data, from a variety of sources, including from actual high stakes money play, that the best players in the world have a cubeless error rate of about .005 points per unforced legal move. At this error rate, you do commit blunders, and many other mistakes at the .05 level. Since these mistakes will reliably show up as mistakes in rollouts, I disagree that an expert can't get much value from using JF level 7. Even the best players in the world could benefit from studying using JF evaluations alone. If you just ignore the errors below .04, you can get feedback without doing any rollouts, and the great majority of JF's suggestions will be correct. The above pertains to checker play. For cube actions, an expert player really wants rollouts, since the absolute error in evaluation is now significant. However, for a player who can't maintain a 1750 rating, I still think they are just as well off studying evaluations. > Perhaps you are more patient and more willing to do the work needed to > get the most out of JF. I like to play a match against a top human > player online, import it to Snowie, and then have Snowie do a 3888 > rollout of every play in the match (which takes ovenight on a P-Pro 200) > while I sleep. This method yields excellent data for just a minute or > two of work. It becomes easy to review the entire match or just to > focus on the rolls where there was an error. > > How long would it take to do the same in JF? More than 10 times longer > (perhaps 50) and more time than I have the patience for. Hmmmm... Yeah, if you are going to do rollouts of the whole match, Snowie may have that 50:1 time advantage. Its huge, whatever it is. Depending on the hardware you have and your skill level and how much you play, this is a big advantage for Snowie. I mostly look at evaluations, because that way I can get feedback on many matches quickly. I just rollout what I consider to be the 'interesting' positions. The amount of work to get the data with JF is many times more than with SW, but if you include the time that it takes to actually absorb the data (yeah, studying) then it's not as big a deal. This time cost doesn't change. > This is an extraordinarily high quality release 1.0, better than any > other software product that I have ever seen and I doubt that anyone > can point out counterexamples. I agree that Snowie looks to be a great product. I'm not sure to whom I would recommend it at $599, but at $299 it will be worth it to anyone who thought the JF analyzer was worth it. I'm not sure it's worth $299 to a more casual player, someone who doesn't enter tournaments or play for money. It certainly isn't worth $599 to them. David Montgomery monty@cs.umd.edu monty on FIBS

Did you find the information in this article useful?

Do you have any comments you'd like to add?