Getting started with GNUBG – A Guided Tour Part 2
by Robert-Jan Veldhuizen

Match Analysis
One of the most useful things you can do with GNUBG is to have it analyze backgammon matches you played, or interesting matches from others. If you play on-line, then it's simply a matter of saving the match from your backgammon client in a format GNUBG understands, importing it with GNUBG, let GNUBG analyze it and then browse and save the results or export your match in the form of a web page, etc.

If you want to enter matches from "live" play or from a book, you will need to enter moves and dice rolls manually into GNUBG. You can also set up single positions that interest you; for that you need to know how to set up board positions and match information. These topics will be covered in another section.

Saving your on-line matches
Nowadays, there are many different backgammon servers (sites) on the Internet where you can play backgammon. For now, I will just describe how to get matches from FIBS, GamesGrid and TrueMoneyGames into GNUBG. Procedures for other servers are probably very similar. As long as you can save your backgammon matches in a suitable format, you should be okay.

GNUBG understands the following formats for matches or money sessions:

  • *.mat (Jellyfish match)
  • *.sgg (GamesGrid / Snowie)
  • *.tmg (TrueMoneyGames / Snowie)
  • *.bkg (BKG session / Berliner)
  • *.txt (FIBS oldmoves)
There are several different FIBS clients available and it's even possible to play directly by telnet. Most clients will automatically record your matches while you play them, but you will have to remember to save them as *.mat files (BBGT),

missing image!

or convert them to those (JavaFIBS, 3DFiBs). If you play by telnet, use the oldmoves command on FIBS and save the output as a simple text file.

Check the "Record" button to make sure the program is set to record your matches (on by default). Save them as *.sgg files.

missing image!

The client has an auto-save feature that you may want to use. From the "Options" menu, click "Backgammon settings..." and choose the "Auto-Save" tab. Tick the appropriate boxes.

Just go to the "File" menu, click "Save as..." and save as *.tmg file.

missing image!

To do this automatically, you can use the "Tools" menu from the main TrueMoneyGames window. Goto "Options", click tab "Backgammon", under "Save session": check "Automatic" and choose the *.tmg file type.

Importing your matches
In GNUBG, click on the "Import" button and a window will pop up:

missing image!

Click on the "Import from format" drop-down list and select the appropriate file type. Then, navigate to the folder where your matches are stored in. This depends on your computer set-up and the program you use to play backgammon on-line. For Windows, it is usually a folder located in C:\Program Files\NameOfClient\. For JavaFIBS, look under matches\jellyfish, other programs usually have a SavedGames folder.

missing image!

To navigate through your folders, use ..\ to go to the parent directory (one level up), or use the drop-down list at the top middle to go up several levels at once. At the bottom of the folders list you can click different harddrives, CD/DVD-players, partitions, etc.

missing image!

Once you found the right folder, it is a good idea to let GNUBG remember it. Before loading your match, click on "Set Default Path" to do that. Later, from the main window, use menu "Settings" and click "Save settings" to make this change permanent. GNUBG can remember different locations for each file type.

Analyzing your matches
Let's start by telling GNUBG how and what to analyze. Go to menu "Settings" and click "Analysis...". A big window with many options opens:

missing image!

There are a lot of options there, but fortunately you don't have to bother much with them usually. Here's a few that you may want to change:

  • Move Limit: This is the maximum number of analyzed moves per turn, that GNUBG will display and save. 20 is reasonable, but sometimes not enough. You may want to set this to 35 or higher so you never miss an interesting move in the list, if you don't mind the bigger filesizes this creates.
  • Skill thresholds: After analyzing, GNUBG will specially mark errors larger than these values, so you can find your biggest errors quickly. The default settings should suit most players; strong players may want to lower these values.
  • Checker play and Cube decisions: Here you have many options to set the quality and speed of the analysis. I suggest to use one of the predefined settings. Instead of the default "expert" setting, which analyses at 0-ply only, I recommend using "Supremo" both for checker play and cube decisions. This will analyse at most 16 candidate moves at 2-ply every turn (similar to Snowie's 3-ply "Huge" setting). A very strong setting, yet still quite fast.

When you're done, click "OK" and remember to save settings if you want to make your changes permanent! Now we can start the analysis.

missing image!

Choose "Analyze match" or alternatively, "Analyze session" for money games. You can also analyze single games or even single moves, but this is not very useful for general purpose. A progress indicator will appear at the bottom right:

missing image!

GNUBG will play a broken chord sound when it's finished. Before looking at the results, I suggest saving the match and analysis at this point. This is similar to saving a match you played against GNUBG, just click the "Save" button and enter a name for the SGF file.

Reviewing analyzed matches

The match statistics

Let's start by taking a look at GNUBG's overall verdict. From the "Analyze" menu, click "Match statistics" or "Session statistics". A window with several tabs will open, initially showing the overall statistics:

missing image!

I will give a brief description of the data presented here, row by row:

  • Error rate (total) The amount of equity given up over all games, first in EMG points and in brackets, as a MWC percentage. Lower is better.
  • Error rate (per decision) GNUBG divides the above error figures by the number of unforced moves and reasonably close cube decisions and displays how many millipoints of EMG you gave up per decision. This is a good indicator of your backgammon skills. In brackets, the percentage of MWC given up per decision. Lower is better.
  • Equiv. Snowie error rate Another good indicator of your backgammon skills, similar to the previous number, but based on Snowie's slightly different method which on average gives a figure about twice as low. Lower is better.
  • Overall rating Based on the GNUBG error rate. GNUBG is very harsh in its judgement here, hopefully this will change in a future version. For now, conside yourself one category higher than what GNUBG calls you ;-)
  • Actual result At the start of the match, GNUBG gives both players 50% chance to win. Here is what happened to that initial percentage. In other words, +50% means you won the match, -50% means you lost! This figure is more interesting when you look at incomplete matches or individual games of matches.
  • Luck adjusted result GNUBG also measures the amount of luck a player got every roll. Adding it all up, GNUBG compensates the actual result above for luck and comes up with an adjusted figure. +0.83% and -0.83% here means that GNUBG thinks Y-33 had 50.83% chance to win this match, compared to my 49.17%. Since GNUBG has "removed" the luck, this should be an unbiased skill indicator. Unfortunately, it isn't always very accurate, especially not for shorter matches, but a useful figure nevertheless.
  • Luck based FIBS rating diff. Based on the above percentages and the matchlength, GNUBG gives the corresponding (FIBS) rating difference between the players.
  • Error based abs. FIBS rating GNUBG uses a sophisticated method, looking at checker play errors and cube decision errors per decision seperately, in combination with the matchlength, to estimate how a player would probably be rated on FIBS on average, based on this single match. This is a good indicator of your backgammon skills. Higher is better.
  • Chequerplay errors rating loss By default, GNUBG considers itself rated as 2050 on FIBS. Here is what it subtracts from that value for your checkerplay errors. Lower is better.
  • Cube errors rating loss Ditto for your cube errors. These last two combined give the earlier mentioned estimate of your FIBS rating. Lower is better.

Which of the numbers above are most important or reliable? That is not an easily answered question and subject to debate; there are many factors involved. The most important thing to realize is that GNUBG's data will be more reliable and meaningful when the match or session is longer. Especially the number of moves or decisions is important. Anther point worth considering is that GNUBG (2-ply) is not perfect. In certain gametypes it may be unreliable and therefore produce wrong error rates.

missing image!

Click on the "Checker play" tab to get more details about checker play. You will see that GNUBG keeps track of unforced moves to get a better estimate of how many decisions you actually had to make (dancing on the bar or playing the only possible move doesn't count). The moves marked doubtful, bad and very bad give an impression of the size of your mistakes. The error rates are calculated the same way as in the overall statistics, except that only checker plays are considered, not cube decisions. The rating GNUBG produces is often a little (too) pessimistic, just as with the overall rating.

missing image!

The "Cube decisions" tab produces some more complicated statistics, here's what they mean:

  • Total cube decisions A decision is when you have cube access (center cube or own cube), so you can (re)double or not, or when your opponent offers a double, so you can take or not.
  • Close or actual cube decisions Every (re)double and every take/pass decision is counted as an actual decision, however, the decision whether to (re)double or not is only counted when it's at least somewhat close (within 0.25 EMG).
  • Doubles Actual doubles made
  • Takes Actual takes made
  • Passes Actual passes made
  • Missed doubles (below CP) When the player should have doubled in a position that would have been a take for opponent.
  • Missed doubles (above CP) When the player should have doubled in a position that would have been a pass for opponent.
  • Wrong doubles (below DP) When the player should not have doubled because he wasn't good enough.
  • Wrong doubles (above TG) When the player should not have doubled because he was too good, i.e. he should have played on for the gammon.
  • Wrong takes When you should have passed the cube instead.
  • Wrong passes When you should have taken the cube instead.
  • Error rate (total) All the above errors together
  • Error rate (per cube decision) The above number divided by the number of "close or actual cube decisions".
  • Cube decision rating Based on the previous figure. Here, GNUBG's grading is at its toughest, don't be surprised to get an "Awful!". Hopefully this will be changed in the future.

It should be noted that due to the relatively low number of close or actual cube decisions in an average backgammon game, the error rate per decision and the rating aren't very useful unless you analyze long matches (like 11 points or more) or sessions (at least 10 games or more).

I'll skip the "Luck" statistics as they aren't that useful anyway.

missing image!

GNUBG's "Graph" is useful to get a quick overview of the errors by both players during a match. It shows the total error (EMG) for each player on a per game basis, and for all games combined. The green part represents checker play errors, the blue part cube decision errors. So, here it says that player Y-33 hardly made any errors in the first game, while Zorba made some checker play errors and cube decision errors in roughly equal amounts. In the second game, neither player made any cube errors and Y-33 gave up more with checker play errors than Zorba. Overall, both players were roughly equal with Zorba better at checker play but worse with the cube.

Statistics per game

If you like, you can get almost all of the above statistics for single games of the match as well. To do so, use the navigation buttons at the bottom or the drop down list, just as you would in the game record. The extra button with the red diamond gets back to considering all games of the match together. It should be noted that determining players' skills on the basis of one single game is of course unreliable, but at least you can see which games gave a player the most trouble.

No panels

If you uncheck "Split statistics into panels", you will lose the graph and the the other statistics will just be combined into one long list

Copy results to clipboard

Right-click on the statistics to copy the text from the statistics to the clipboard. You can choose to copy just the current panel view, or copy all panels at once. The graph will not be copied.

This concludes the second part of GNUBG's guided tour. Next we're going to look at the match itself and what GNUBG has to say about the moves and cube decisions made.

Go to Part 3

Copyright © 2006 by and Robert-Jan Veldhuizen

Lists of backgammon tournament winners can be found in the Champions section.

Copyright © 2006-2011 by All rights reserved. This material may not be broadcast, published, re-written, or re-distributed in any form.