Thoughts on Fruit=Rybka EVAL

[BB+]

I might also note that I estimated that it was less than 1% of the issue, so ignoring it doesn't change much.

In the documents the PST's are promoted as a major issue.

You are making the same error ChrisW made, when he tried to promote the same idea. Namely the "issue" to which I referred in the context above (post here) was copyright infringement (on an overall basis), while for copying/originality, PST is more significant, and as others have pointed out, these items do indeed "add up".

Lemme pick just one to comment, because in the end that's the one what this is all about: "I am so sorry for all your work"

I agree that a lot turns on your use this phrase, as I said previously I found to be quite dismissive, both of me, and the Panel's diligence. As I said previously [in the aftermath of the "silly" PM], perhaps something was lost in the translation, but the tone was quite annoying too. Particularly when you then asked me to do more work.

Furthermore, returning to the context of your statement:

Fruit 2.1, Rybka 1.0 Beta, and Rybka 2.3.2a all give a penalty for an isolated pawn that depends on whether the file is half-open or closed.

I do that since 1982/83.
Because,
It's the right way of doing chess knowledge.
Vas is an IM, he knows.
I am so sorry for all your work.

Note that I then pointed out that REBEL did not do the same as Rybka and Fruit (that is, REBEL has a square-based bonus). So I found you to essentially ignore what EVAL_COMP was about, and that you "dismissed" it without understanding it (and why should VR's IM status matter?). If you want to dispute EVAL_COMP on other grounds, that's fine, but currently it is unclear to me that you understand what it is and/or tries to measure.

As it seems to me you had too little opposition in the panel and also overlooked a couple aspects of chess programming. The PST's comes to mind, but also EVAL. And we haven't even really started a thorough discussion about EVAL as we are still discussing the PST. All in all, the documents are inconclusive.

Again you reconstruct the Panel atmosphere in terms of "opposition" (to me?), while its progression did not follow in this manner. I have said before, and will say it again, that the criticisms of evidence (both that presented by me, and by others) from Panel members were notably more substantial than anything that has been produced afterwards.

The idea that I (or Bob, or Zach) rammed through the evidence, and the Panel was merely rubber-stamping, is wrong -- further, you [and others] keep pressing this issue, no matter how many times it is stated otherwise. I don't know whether this is another vestige of trying to make the issue about persons (rather than evidence), or what. Since I've repeatedly made this point to no avail, at this point I would suggest that you ask Panel members like Gerd Isenberg, Don Dailey, and Ken Thompson (all of whom voted, as is evident from the Report) whether there was "too little opposition", and whether they (and others) "overlooked a couple aspects of chess programming". Just as one example, as MarkL has noted there was a "survey" after a few weeks, asking people what they thought, and where the issue stood -- there were 80+ posts in that thread, which led to further avenues of discussion/debate, etc. Other threads were similar (though usually not so voluminous, as each [sub]topic had its own heading).

And you (and Zach as well) should REALLY explain to me that when you state that the EVAL order in Rybka is basically the same as in Fruit that is downplayed (by Zach) as irrelevant. I give you guys a smoking gun and you call it irrelevant? The logic fully escapes me.

I will say (again) that I find the evidence with EVAL order to be quite debatable either way. If the evidence were clear, then I agree it would be much more relevant. I think this is in congruence with the Panel behaviour -- we threw out various pieces (e.g., 10-30-60-100 scaling) on a similar consideration, that they were either too dubious and/or contestable/mitigatable. OTOH, the "root search ordering" (Section 4 of RYBKA_FRUIT) is more clear, and this is why it was left in [though it still wasn't mentioned too much, as it seems much less relevant than other issues in the end].

[BB+]

What's faster?
midgame=midgame+(x*11); endgame=endgame+(x*14); // fruit way
versus,
if (phase) midgame=midgame+(x*11); else endgame=endgame+(x*14); // classic way

Branching is slow with mis-predictions, but likely you would predict correctly almost all the time (being wrong only at the phase junction). Another possibility would be to try to use cmov (conditional move) somehow, though see below for what I think is the "real" solution (at least on 32-bit machines -- not sure about the "classic" 8-bit processors).

In any event, Fruit 2.1 interpolates between midgame and endgame, essentially on a 0-to-24 basis. As such, both midgame/endgame are computed for every element, and only at the end is there something like

(midgame*phase+endgame*(24-phase))/24

used to glue these together. This interpolation is not unique to Fruit, but I think Fruit did do a good job of demonstrating the value of such an approach (e.g., it smooths any "jumps" when passing from phase to the next). [Note: Fruit 2.1 uses linear interpolation, while Rybka 1.0 Beta has a more complicated interpolation scheme. Later Rybka versions remove any complication here, and have linear interpolation].

In later Rybka versions, the "score[2]" array is turned into a 32-bit packing of two 16-bit entries into one variable, which solves all problems (you multiply x by, say, (11<<16)+14).

[Rebel]

Second, you really blew it with "A lot of the evidence hasn't even been written up properly". That turns into "A lot of the evidence is improper and worthless."

I agree with you.

The impossible just happened

Zach's honesty is a gift from heaven for any lawyer.

[Rebel]

I might also note that I estimated that it was less than 1% of the issue, so ignoring it doesn't change much.

In the documents the PST's are promoted as a major issue.

You are making the same error ChrisW made, when he tried to promote the same idea. Namely the "issue" to which I referred in the context above (post here) was copyright infringement (on an overall basis), while for copying/originality, PST is more significant, and as others have pointed out, these items do indeed "add up".

A lawyer with Rybka glasses will read that different. Seriously, check your own words. Here we can talk reasonable although being opponents, in court it's war and perception is all.

[Rebel]

Lemme pick just one to comment, because in the end that's the one what this is all about: "I am so sorry for all your work"

I agree that a lot turns on your use this phrase, as I said previously I found to be quite dismissive, both of me, and the Panel's diligence. As I said previously [in the aftermath of the "silly" PM], perhaps something was lost in the translation, but the tone was quite annoying too. Particularly when you then asked me to do more work.

Sorry for doing it again in the above one. Don't read it as something personal, it's not meant that way. It's my passion for the subject that drives me.

[Rebel]

Furthermore, returning to the context of your statement:

Fruit 2.1, Rybka 1.0 Beta, and Rybka 2.3.2a all give a penalty for an isolated pawn that depends on whether the file is half-open or closed.

I do that since 1982/83.
Because,
It's the right way of doing chess knowledge.
Vas is an IM, he knows.
I am so sorry for all your work.

Note that I then pointed out that REBEL did not do the same as Rybka and Fruit (that is, REBEL has a square-based bonus). So I found you to essentially ignore what EVAL_COMP was about, and that you "dismissed" it without understanding it (and why should VR's IM status matter?). If you want to dispute EVAL_COMP on other grounds, that's fine, but currently it is unclear to me that you understand what it is and/or tries to measure.

Rebel HAS a penalty for an isolated pawn that depends on whether the file is half-open or closed.

But you went to my website, did not find the information and concluded Ed is wrong while missing what the page also said,

This ends the project to describe REBEL in pseudo code. I estimate that what's written on this page is responsible for 95% of REBEL's play, to describe the remaining lesser dominant stuff (many small adjustments and exception code, hardly responsible for REBEL's main play) fall outside the realm of this project and would require a separate page of a similar size.

[BB+]

A lawyer with Rybka glasses will read that different. Seriously, check your own words. Here we can talk reasonable although being opponents, in court it's war and perception is all.

Here is the quotation:

[...] If there were nothing but the PST evidence for Fruit/Rybka, I might say that [if successful in proving this] FL could could claim some piddly (<1%, perhaps <0.1%) amount of Rybka profitability under the heading of copyright infringement. OTOH, I would continue to insist that such "piddly" amounts are indeed "actionable" under the law [...]

This is from the thread to which I referred ChrisW in the context you quoted. I think the meaning of "1%" being copyright infringement is rather clear.

Is quoting things out of context likely to get a reprimand (or more) from a judge? Usually in a court situation, clarifications and/or "redirects" are allowed, indeed, even encouraged. [Perhaps the most ridiculous contextual mis-step was, well, over the word "ridiculous"].

E.g., the snipping of Zach's comment again seems to have more context:

In retrospect, I regret only publishing evidence relating to the evaluation, since it's fuzzy in many ways, and people can easily see it as "circumstantial". A lot of the evidence hasn't even been written up properly. I view this as my own failure in many ways, but there's only so much effort I'm willing to put into this.

Ultimately, what I mean is that, if there are people who are capable of understanding the evidence, but still disagree with me, it is partially my own fault for not making a convincing enough case.

Again it seems that taking the second sentence of the first quotation would be misleading (as Bob did, to poke fun at the "Rybkarians"), at best.

[BB+]

Rebel HAS a penalty for an isolated pawn that depends on whether the file is half-open or closed.
But you went to my website, did not find the information and concluded Ed is wrong while missing what the page also said, [...]

Then indeed, we really were talking past each other. I wasn't even commenting on whether REBEL did open/closed or not. I was only emphasising the square-based nature of the REBEL penalty (which seems to be listed in a PDF redaction, but not your website AFAIK) as being different from Rybka/Fruit/(Faile), and saying that this in itself was different from they did.

I am sorry if you thought that I concluded you were wrong about open/closed, but we were simply talking about two different things. My exact words were:

Your write-up in How REBEL Plays Chess suggests that, at least at some point, you did otherwise than Fruit/Rybka (which do not depend on the square). [...] [then it listed the square-based diagram]

So I tried to emphasize the "square-based" nature as being the "point of difference", but perhaps we were just on different wavelengths.

In short, each EVAL_COMP "feature" has many conditions, and I only commented on one of these (the difference compared to Fruit/Rybka [and indeed almost all others] in being square-based, which I was able to determine), while you referred to another condition (the open/closed similarity to Fruit/Rybka/Faile, which I obviously didn't know about, as you say). So maybe it's not true (for EVAL_COMP purposes, which need quite a specific description) that "But you can take Rebel, most of the eval stuff is an easy read" is valid after all?

[Rebel]

Again you reconstruct the Panel atmosphere in terms of "opposition" (to me?), while its progression did not follow in this manner. I have said before, and will say it again, that the criticisms of evidence (both that presented by me, and by others) from Panel members were notably more substantial than anything that has been produced afterwards.

The idea that I (or Bob, or Zach) rammed through the evidence, and the Panel was merely rubber-stamping, is wrong -- further, you [and others] keep pressing this issue, no matter how many times it is stated otherwise. I don't know whether this is another vestige of trying to make the issue about persons (rather than evidence), or what. Since I've repeatedly made this point to no avail, at this point I would suggest that you ask Panel members like Gerd Isenberg, Don Dailey, and Ken Thompson (all of whom voted, as is evident from the Report) whether there was "too little opposition", and whether they (and others) "overlooked a couple aspects of chess programming". Just as one example, as MarkL has noted there was a "survey" after a few weeks, asking people what they thought, and where the issue stood -- there were 80+ posts in that thread, which led to further avenues of discussion/debate, etc. Other threads were similar (though usually not so voluminous, as each [sub]topic had its own heading).

Why take (almost) everything I say so personal? Why this defensive attitude?

It's a fact the PST's are under pressure, and I am being mild here. We are talking about PST's over a month by now. And still no consensus. And by your own admission the PST's were hardly an issue for the panel. Not enough opposition over there. All facts. An admission of the facts would be fine.

No need to for you to reply creating a false sentiment that I have no respect for people like Gerd Isenberg, Don Dailey, and Ken Thompson.

Don't try Hyatt tactics one me.

I respect your work too much for that.

Just the facts.

[Rebel]

And you (and Zach as well) should REALLY explain to me that when you state that the EVAL order in Rybka is basically the same as in Fruit that is downplayed (by Zach) as irrelevant. I give you guys a smoking gun and you call it irrelevant? The logic fully escapes me.

I will say (again) that I find the evidence with EVAL order to be quite debatable either way. If the evidence were clear, then I agree it would be much more relevant.

But the evidence is clear and not debatable at all. You and Zach have stated Fruit EVAL order is VERY similar to Rybka EVAL order.