[1eyedjack]

Many thanks, Stephen. I suspect from what you say that even I might have managed slightly better if I was not in such a hurry.

One point that confuses me, being a non-techie person, is that I would expect the program to have to examine every hand in the database once in order to reject it as falling outside of the search criteria. It may be obvious to others but I don't understand how narrowing the search criteria will get around that problem. Does the time saving come within the process of comparing a selected hand with the search criteria? That would surprise me.

As to your objection regarding the usefulness of the outcome, I agree with your and Ben's criticisms entirely. I was hoping that I could do a "search within results" secondary search after the initial one, to test the effect on the overall stats of tweaking the criteria. Of particular interest to me would have been the specific (in order) 4-2-2-5 shape (where there are no rebid problems in a natural system by opening 1C). At least the raw result might be of value to some (and they do exist) who religiously will NEVER open 1NT with two doubletons.

[Rossoneri]

I probably got the same problem as jack did.

Was trying to look at transferring or not with 5332 hands, and ended up trying to run 2 searches on 2 different computers - desktop searching 1NT AP hands and laptop searching 1NT-2D/2H hands. After 1 hour, I gave up and went to bed.

Was out since morning so I didn't get to try again....

I only finished the 1st video, so maybe I will look at the others and try to figure out...I quite like the idea of getting so much statistical data to look at.

Preliminarily, it seems to be that not transferring is the better option. But of course, my search could very well easily have been biased...

[PassedOut]

My situation is different in that I'm very familiar with computers and databases. Because of a late start, I did not get to try out all the features that I'd have liked, but I certainly know now that I will be a customer.

The videos were helpful to me, and without watching them I would have had a tougher time getting going.

The densely packed main screen is the sort of thing one really appreciates once a powerful program becomes familiar, but it does look daunting at first. That's always a problem with UI design, as Alan Cooper has pointed out often. And I'm sure that's why Stephen created the wizard mode.

I'm looking forward to checking out more and more of the capabilities of BridgeBrowser. Thanks for making the trial version available yesterday!

[sfbp]

1eyedjack, on Mar 9 2008, 12:55 AM, said:

Many thanks, Stephen.  I suspect from what you say that even I might have managed slightly better if I was not in such a hurry.

One point that confuses me, being a non-techie person, is that I would expect the program to have to examine every hand in the database once in order to reject it as falling outside of the search criteria.  It may be obvious to others but I don't understand how narrowing the search criteria will get around that problem.  Does the time saving come within the process of comparing a selected hand with the search criteria?  That would surprise me.

As to your objection regarding the usefulness of the outcome, I agree with your and Ben's criticisms entirely.  I was hoping that I could do a "search within results" secondary search after the initial one, to test the effect on the overall stats of tweaking the criteria.  Of particular interest to me would have been the specific (in order) 4-2-2-5 shape (where there are no rebid problems in a natural system by opening 1C).  At least the raw result might be of value to some (and they do exist) who religiously will NEVER open 1NT with two doubletons.

Well, the good news is, I ran the search without doing it via the net. Ie I took a real dataset on a real machine and let it go full tilt!

What a difference! Now it takes 5-10 minutes. Sorry I didn't watch but I had other fires to fight, notably the clock change.

The downside (of course) is that it takes about 1/2 of the CPU.

Here it is:



I'm sure someone will want to give detailed commentary on this (hint hint) but I will content myself by saying this: the SD (actually SEM) is available but not shown in this graphic. The SEM for low frequency actions is large. The SEM for high frequencies is pretty small.... ergo, don't rely on the number you see for 2C but DO rely on 1NT.

I wish I could anticipate all the things that people might do, and create easier ways to access the data. The (perpetual) problem is that to be really interesting, people feel they have to look at the data in some way noone did before. After all if the answer is known, why bother?

So if you are serious about the 5422 thing, the only real solution is to put you and about a gig of data in the same room. It's bad enough being over the net, and in England at that (whereas I am only about 50ms from the server and awm is even closer, about 20ms!!!). This is the way we did searches at the beginning, I may add, until I came up with all these weird ways to index data. Originally I thought "maybe this will never be useful remotely". But you can get a lot of information (drilling down, they call it nowadays) on very specific things, like specific players who bid X holding Y or Z. Saving an intermediate file is definitely possible but it would take way too much CPU and disk on a shared server. As you've seen many ideas take a scan of almost the entire data set, and no one else gets a look in.

Ben will have the answers to these and many other questions over the coming months.

Stephen

[inquiry]

Well, not everyone will have watched the video on bid analysis on homebase's page, nor taken the six lessons in the bridgebrowser university on the same pages. So for them, the image Stephen posted will be a little difficult to follow. So let me see if I can explain what stephen showed...

First, he did a seach for all hands were dealer had precisely 5422 distribution (any suit could be 5 or 4 or 2) and 15 to 17 hcp. There were 51,498 of these. He forced opener to bid something (no passes in teh table) so he found auctions for 51,276 hands. Lets examine the top of image...





I have highlighted two areas in red. Inside the red circle is an arrow shaped thingee pointing to the column header "open". That means the "data" (imp average and mp average) will apply to the opening bid. Inside the long red box we see the following items...

1C 8299 0 0 0 0 0 81 0 -0.12 4792 48.30 3507

What this means is 1C was opened 8,299 times, and after an opening bid, no one bid one club as overcaller, responder, advancer, opner rebid, or made later in the auction (all such bids would be insufficient of course, which is why). One club was the final contract by the OPENING side 81 times, and 1C was never the final contract by the non-opening side. Of the 8299 hands that were opened 1C, the average result was a minus 0.12 imps (4792 opening 1C bids at imps), and the average mp score was 48.30% (3,507 opening 1C bids at MP).

If you were to move the sliding arrow (in the red circle) over the column headed by CntOS (for contract opening side), the average imps and matchpoints columns would reflect what the results were for the 81 contracts played in 1!C. And the number of matchpoint and imp hands would change from 4792 and 3507 to a pair of numbers that add up to 81 (for the 81 hands played in 1!C).

Now, lets look at another part of the image....here i placed a green box around the opening bids of 1!C, 1!D, 1!H, 1!S, 1NT, 2!C, 2!D, 2!h and 2!S. One might understand a 2!C precision opening with 4M5!C and 15 hcp. 2!D through 2!S might be a two suited opening bid. I know ritong (and I) will open 2!S or 2!H with five in that major and 4!C on up to 15 or maybe 16 hcp. In the database, you could click on the 13 hands that opened 2!S and see the hands and the alerts, if you were curious.



But if we look across the table, we find that an opening bid of 1!C, 1!D, and 1!H all averaged below 50% MP and negative imps. The 1!S opening bid averaged essentially zero imps (slight negative) and slightly more than average MPs. While 1NT averaged better very plus imps and MP's.

This search was set up with very broad limits.. it could include, for instance xx in both doubletons, one could fine tune it, making Qx or better in doubletons a requirement if one wanted to.

In addition, you can use the plotting routines to see if 15 versus 17 hcp made a difference in any of the bids, you could look to see if opening 1M worked better with weak major or strong major. All that from this screen without doing any more searching (the right click on the bids and plotting is a powerful tool).

At the very least, this study suggest that opening 1NT with 5422 distribution is not a losing proposition, and might very well be a winning one. To fine tune your decision process into rather to open 1NT with such distibution, you might want to examine the type of hands where 1NT turned out poorly. That is easy enough, right click on 1NT and choose plot. Look for contracts played, or tricks won, etc.. to examine how typical auctions might go.

Pop quiz ... from the table above, how many times did opener rebid 1NT after opening something other than 1NT?

[1eyedjack]

Deleted post - rubbish

[barmar]

1eyedjack, on Mar 9 2008, 04:55 AM, said:

One point that confuses me, being a non-techie person, is that I would expect the program to have to examine every hand in the database once in order to reject it as falling outside of the search criteria. It may be obvious to others but I don't understand how narrowing the search criteria will get around that problem. Does the time saving come within the process of comparing a selected hand with the search criteria? That would surprise me.

If any of the criteria you use when narrowing is indexed, the search will be much faster.

To the non-techies, an index in a database is similar to an index in a book. Imagine you've got a history book, and wanted to find everything it says about George Washington. Without an index, you'd have to read the entire book, which would take a long time. But with an index, you can go directly to the relevant pages and it takes almost no effort.

Similarly, if any of the criteria in your BridgeBrowser search is indexed, brbr can just look at those relevant hands, and compare them against the remaining criteria, rather than having to check every hand in the database. Thus, searching for hands fitting critera A+B can be faster than searching for A if B makes use of an indexed feature; for instance, if player names are indexed, searching for 5422 hands held by Dano would be much faster than searching for all 5422 hands.

To relate this to the history book analogy, suppose "horse" isn't in the index. If you wanted to find all mentions of horses, you'd have to read the entire book. But if you wanted to know about George Washington's horse, you'd start by looking up George Washington in the index, then check those pages to see if his horse is mentioned, which would be a hundred times faster.

[1eyedjack]

Barmar, that's a great description, thank you.

Now, a suggestion to Stephen, if you are still monitoring this thread:

An idea that you may wish to consider is to allow permanent free online access to a very small sample database, available to any non-subscriber. Any statistical value drawn from it would be worthless (indeed you could make the data entirely ficticious if so minded), but the sole purpose would be to allow a potential customer an opportunity to evaluate the software at his own pace, make his own mistakes, and learn from them, until he gets to the point where he feels that he would derive positive value from subscribing to the full-blown product.

To some extent he may fall into the trap of running inefficient searches without appreciating their inefficiency - the small size of the database would prevent a sufficent effluxion of time running the search for the user to appreciate the flaw in approach. BUT, this would allow the user a good opportunity to familiarise himself with the interface.

I say this, having just had a thought (which didn't work out): I know that the free trial is over, during which short period I learned next to nothing, so I figured I will just run the program anyway, not expecting access to the data, but just so that I could have a good look through the menu structure and help files. You really have to look at what happens if you try this without entering a valid password. You get into all sorts of endless loops and error messages even clicking on the cancel key.

[inquiry]

1eyedjack, on Mar 11 2008, 02:52 PM, said:

Barmar, that's a great description, thank you.

Now, a suggestion to Stephen, if you are still monitoring this thread:

An idea that you may wish to consider is to allow permanent free online access to a very small sample database, available to any non-subscriber. Any statistical value drawn from it would be worthless (indeed you could make the data entirely ficticious if so minded), but the sole purpose would be to allow a potential customer an opportunity to evaluate the software at his own pace, make his own mistakes, and learn from them, until he gets to the point where he feels that he would derive positive value from subscribing to the full-blown product.

To some extent he may fall into the trap of running inefficient searches without appreciating their inefficiency - the small size of the database would prevent a sufficent effluxion of time running the search for the user to appreciate the flaw in approach. BUT, this would allow the user a good opportunity to familiarise himself with the interface.

I say this, having just had a thought (which didn't work out): I know that the free trial is over, during which short period I learned next to nothing, so I figured I will just run the program anyway, not expecting access to the data, but just so that I could have a good look through the menu structure and help files. You really have to look at what happens if you try this without entering a valid password. You get into all sorts of endless loops and error messages even clicking on the cancel key.

First Jack, stephen already allows free access to a small database, the homebase one, with the really only "requirement" being registered at the homebase forum. Send me a message over there (after you register) and I will give you the user name and password (in theory, there is a requirement to play at least once in any homebase events, but that is simply too difficult to keep up with... but we will keep that part our little secret, be sure not to tell anyone).

Second, Stephen has agreed to add millions of hands to the homebase database (or pull out a single several million hand database) to serve exactly the purpose you suggest. This will not be "fake data", and I hope it will be all BBO data rather OKbridge data. That database will again be free, probably in theory requiring registration at homebase as well.

Third, Stephen has committed to allowing totally free access to the full database (over 250 million hands) to regulars of homebase events. The definition of regulars is what is still being discussed. The thoughts seem to be around 20 plays within several month period, something like that.

I can not discuss the program further in this thread.... but send me a pm on the homebase forum, and I will get you all set up to access now the current small database, and when the larger free one is available, I will get the info for that ASAP.

[sfbp]

1eyedjack, on Mar 11 2008, 11:52 AM, said:

You really have to look at what happens if you try this without entering a valid password.  You get into all sorts of endless loops and error messages even clicking on the cancel key.

I just tried it here. No problems, I tried every combo I could see, given that I entered a bad password.

I decline to bore everyone with the details, for obvious reasons. The first choice you had was a Yes/No. If you click N(o, the program exits immediately. If you click Y(es you get another opportunity to enter a username and password.

Granted, if you start the program without using the shortcut (by double clicking it) you will get lots of funny errors, but that goes for most software on the market. There's a built-in assumption that you know what you are doing, and have a local BRBR installation. This can be useful when the net is down, but that mode was not designed for someone who is locked out. As Ben says, installing hbrbr instead (de-install xbrbr, first) would achieve your current objectives.

Something else you might consider is simply to open the windows help file BRBR.HLP, this time by double-clicking it.

That would give you lots of reading