GNU Scrapbook of Problems

[sixty_something]

when i first started using GNU, dorbel planted a seed of curiosity by mentioning collections of positions .. i don't recall whether he called it a scrapbook or whether or not it was even in reference to GNU, but it's the word i've settled on for something i'm now doing .. so, thanks again, dorbel

having reviewed several hundred games in GNU many a move at a time, i'm beginning to recognize the obvious that hitting the highlighted marked moves is perhaps the most efficient use of my time .. reviewing a whole match has it's place, but certain patterns where i or my opponent repeatedly make similar mistakes keep repeating themselves .. it's these i'm putting in this scrapbook

i'm collecting these as saved positions in a separate folder within GNU, gnubg\saved_positions .. the saved position option in GNU's save feature simply saves the snapshot of the dice rolled or cube decision in the context of the match, but without all the other moves .. each of the moves or cube decisions i've saved is one GNU has marked doubtful, bad, or very bad for me or my playing partner using the default threshold settings, 0.04. 0.08. 0.16 .. i'm not keeping track of the original decision, to do so might be rather embarrassing at times .. but my threshold for selecting a move is whether having looked at it, i am still curious about or surprised at GNU's decision and not sure whether i'd make that pick myself next time i see something like it .. most call for counter-intuitive solutions

when reloading this saved position, i began thinking it would be nice if they could be assembled and used as sets of problems to solve .. by turning GNU's tutorial mode on and tweaking the settings a bit before i save the position, this becomes quite possible .. when each position is opened, it is by definition a problem waiting to be solved by making a move or cube decision .. no instructions are otherwise necessary .. with tutorial on GNU's analysis provides it's own built in answer key .. a best move is always picked by GNU and your pick, if not the best, can be scored based on equity error or wasted equity .. for each position, GNU evaluates the decision in tutor mode and brings up the tutorial hint window if the selected decision would be marked .. i've tweaked the threshold setting for doubtful moves from 0.04 as a lower bound to 0.01 .. so, almost anything except GNU's best choice will open the tutorial window

while this isn't much in the way of a fancy interface, it's an easy way to collect positions that make interesting problems for further study, learning, and perhaps sharing

trying to keep them organized and useable in sets is the next challenge .. for now i'm using a descriptive prefix before the imported or saved default file name .. some of my prefixes are:

AggressiveOrSafe_ - so often GNU teaches to aggressively go for the hit, but not always
Backgame_ - timing and position issues from backgame settings
BackMen_ - how to handle checkers left behind, when to break, when to run
BarPoint_ - making the bar point or taking the other option
ConventionalOrNot_ - moves like 61, 31, 42, 53, 65 where conventional options are called into question
CubeEarly_ - cubes thrown early in a game
EndGame_ - positions typical of endgames
FivePrime_ - building, holding, or busting a 5-prime
OpeningResponses_ - not openings, but responses to openings
PlayingDoubles_ - perhaps the most common marked move, playing doubles in many different situations
Redouble_ - redoubling always increases the tension
RunOrCover_ - sometimes it's better to run than seek cover
RunOrStay_ - sometimes it's better to stay than to run
TakePass_ - cube decisions to take or pass
TooGoodToCube_ - another common cube mistake not always obvious
UnluckyRoll_ - minimizing the damage from an unlucky roll can be quite challenging

thus, file names become a little more meaningful and are easily kept unique and include a reference a source file .. the resulting saved position files are tiny, less than 1K each, with names like these:

AgressiveOrSafe_You_vs_teyakis_1189190729000.sgf
Backgame_You_vs_Anterico_1189187491062.sgf
Backgame2_You_vs_Anterico_1189187491062.sgf
BigBlunder_You_vs_minki_1189176020656.sgf
UnluckyRoll_You_vs_teyakis_1189190729000.sgf

these are just an early attempt at categorizing game positions .. on occasion i've named certain positions with 2 or even 3 of the above prefixes, but generally it's an attempt to create some simple order in what has already become 50 plus positions just from my games in F3's yesterday

naturally, given my level of enjoyment in using GNU and trying to communicate it, i've got several other ideas on this .. but i'm going to put this out and see what comes of my own use before reporting back

if you'd like a sample of problems, i'll send you a set of 10 with suggestions on how to grade them, all using nothing but GNU .. better still send me a set of your own .. try it, Mikey, you might like it

sixty

[burper]

rather than using a prefix, which confines you to one attribute (or really long names ,
why not use tags where you could have any number?

[sixty_something]

yes, these file names are a kludge selected because i wanted to categorize the problems .. the additional info from the original file name is just becoming messy, but is an easy way to assure a unique name while saving positions in a match .. i intend to replace it with xxxxxxxnn, where xxxxxxx is the category name, nn just a serial number providing a unique file name .. but i'm doing that by renaming it outside of GNU .. trying to find the next serial number while first saving a position is cumbersome

i'm finding the categories are sufficient, for now .. seldom do i save a position under two categories other than PlayingDoubles which comes up in all categories, but i'm considering folding those back into the other categories eliminating PlayingDoubles as a collected set of positions .. one of the more interesting asides of this collection has been the evolution of category names .. they are becoming almost a checklist for consideration before making a move or chapter headings targeting my weaknesses

so far, i've collected a over 130 positions .. i'm collecting far more than reviewing, as many are still too fresh in my mind .. so, on review i'm too often recognizing the best play from memory, not evaluation - that's no good .. in a week or two, i'll try again, but this looks like a good way to collect interesting positions

what would really be nice is to have an automated grading system, tracking the error rate of answers for each position .. grading is simply by equity error .. repetitive use by me would tend to highlight my repeated mistakes .. use by a large number of users would over time cause the toughest problems to sort to the top of a list

but whether these are tough or the best problems or not, it seems like a good way to jump into thinking about game positions .. it's also quite interesting to play out the position for a roll or two when GNU's recommended decision suggests something that makes no sense at first glance

so, even though this scrapbook isn't very pretty or as automated as i'd like it to be, it is looking useful

[sixty_something]

here's an update on this little learning experiment of mine for anyone interested

i'm still liking this idea, but i've retreated from my original plan of categorizing moves and cube decisions .. i've archived those for possible use later .. after a long chat about this with donzeamon, it dawned on me that what i'm really trying to do is improve MY decisions GNU marks as bad, not create some comprehensive collection of challenging positions, as interesting as that goal may be  .. so, rather than collecting those marked doubtful and bad (mine and others), i'm now just saving my own decisions marked *very bad*

my theory is these are uniquely MY VERY BAD decisions made in fibs matches and when playing GNU .. these decisions each cost me -0.14 or more in equity, some much much more .. for now, i'm pretty much ignoring bad moves made by my opponents .. my thinking is that by dwelling on bad decisions i didn't make, i may even be reinforcing bad habits .. i have enough of my own to worry about .. i mean, if i didn't make the bad choice in the first place, is it really an efficient use of my time to so carefully review them?

of course, it is to a degree, but, if through later review, i can teach myself not to make these and similar *very bad* decisions of MINE the next time, i'm more directly targeting MY unique learning experience at the point where my equity loses are the highest .. this seems much more efficient and is much more manageable

it's too early to tell anything from review since they're still too fresh in my mind .. but already, i've noticed a drop in the raw number of blunders (decisions marked very bad) in matches i review with GNU .. maybe i'm just teaching myself to pay closer attention .. maybe i've just had a couple of good days .. regardless, i'm sure i'm learning something from continuing to review matches with GNU .. time will tell .. but today, tenuously cresting 1700 again, i'm a very happy fibster

[blitzxz]

I'm about to start collecting my blunders too. Up 'til now i have just analyzed my games. But is it somehow possible to but all the positions to just one file (like a collection of positions)? And another question to gnu experts: What is that relational database? It doesn't seem to work.

[sixty_something]

blitzxz, the thought has occurred to me to try saving only the Position ID and Match ID the next time i begin such a collection .. the ID couplet is all required to review or study any position, as those ID's create a unique code including position and match score .. so, a simple flat file, optionally indexed or even annotated, is all that would be required to save them .. this may be the direct answer to your question of how to put "all the positions [in] just one file (like a collection of positions)"

but revisiting this thread again has me thinking - always a dangerous and often a wordy thing

i've now saved over 200 positions gnu marked as "very bad" moves or decisions in a second collection without attempting to categorize them .. i used gnu to save those positions as *.sgf (gnu saved game files), but going back and reviewing them has been tedious at best .. so, i've been less than satisfied with the results .. i've now stopped the collection and review of them (hmmm, another idea to consider restarting)

for me, the next big step is how to use such a collection effectively (whether saved files or ID's) .. cutting and pasting ID's manually is just another tedious step that makes use difficult, at least for me .. while saved files are not as efficient as a flat file of saved ID's, sgf files eliminate the cutting and pasting step and require little more than an ascending non-repeating serial number suffix in the file name to save it .. but therein is the tedium of saving and loading for review - cut and paste ID's (twice) to save and load an ID or name, number, save, and load a file .. it may sound simple, but it becomes mind numbing, for me 

i'd love to have a way to automatically launch gnu to load such sets of my saved positions from either files or lists of Position/Match ID's .. if i had nothing more to do than launch sets of 5-10 problems at a time for review, i could deal with that and even use it daily .. for me, the repetitive and irrelevant task of loading files disrupts my focus and learning opportunity so much, i've found the use of these collections far less than satisfying

yes, i'm innately lazy, especially when it comes to a repetitive task, but the discipline of saving positions marked *very bad* that were more than just dumb blunders has, IMHO, been a very important part of my learning approach .. i'm convinced reviewing such sets could be even more beneficial as i can think of few better ways to focus my learning than toward positions with which i've already had trouble

ideally, storing and tabulating results from reviewing such saved problem sets to see if i am improving would be the ultimate goal of such an exercise .. if, over time, i solved a problem by simply selecting a *doubtful* or even just *bad* choice three times in a row when presented with the problem, i'd be quite happy and ready to tag it as "almost solved" - pushing it to a lower priority subject to less frequent subsequent review

remember, these problems all started as *very bad* choices in real game situations .. i've found through careful analysis that my average equity loss for *very bad* choices is about -0.40 .. that's huge .. merely *doubtful* or *bad* (-0.04 to -0.16 equity loss) is a substantial improvement

if i nailed a problem three times running over time with an unmarked solution (< 0.04 equity loss), i'd tag it "solved" and add it to a set of personal reference positions .. time, btw, is a critical factor in using such problem sets .. i've found it takes weeks, even months, to clear a difficult position from my memory such that i'm approaching it without the helpful hint of familiarity .. it is the familiarity of understanding i'm seeking, not just the familiarity of a puzzle set

this has become my quest for a set of personalized problems and reference positions .. indeed, it has become my Holy Grail for my next big push at learning this confounded, often perplexing, and sometimes beautiful game

i suspect this would be quite doable by using gnu's Python shell for loading problem sets and a simple semi-automatic reporting of the result of each solution in terms of lost equity when finished .. unfortunately, i haven't learned to speak Python yet or convince anyone that does to help me with this approach
any snake charmers interested in lending a hand in this quest? 

[socksey]

any snake charmers interested in lending a hand in this quest? 

socksey



'But you have no chocolate? My dear, how will you ever manage?!?!' - Marquise de Sevigne

[don]

...or, sixty, you could convince webby to add a feature that would let you enter the match and position gnumbers here and have 'em display graphically...

It'd be sweet.
--
don

[webrunner]

...or, sixty, you could convince webby to add a feature that would let you enter the match and position gnumbers here and have 'em display graphically...

It'd be sweet.
--
don

If someone could build that for me i would let them add it to FB.

[KongDan]

[BTW I may post a new thread with this post in, if this thread is dead.]

I've been planning on considerably improving my approach to the game, and spending time on each aspect of the game in turn, and so systemising errors/positions into specific aspects has been a very useful, yet also a tiresome process.

Looking through the various websites on backgammon strategy, I came across this page: http://www.gammonluck.com/nordicseminar.html

I've put each of these aspects [about 40 of them in total] in OneNote with a second column, where I copy links of GNUBG positions [which have may have been edited/altered/rolled out], the links of gammonluck's positions, backgammon forum positions, and screenshots from backgammon books. I have added a few rows to this table to add aspects from Magriel's book, and other aspects that I'm working on. Here is a small screenshot of the first couple of columns


Hope to hear of some feedback of this way of learning

KongDan out

[my FIBS user: SnakeByte]
[my skype id: danimal1993]