Comments on: CBO: the formula for the “density” column statistic (densities part II)

By: density estimate

density estimate — Fri, 02 Apr 2010 19:07:40 +0000

[...] bone reduces resulting in osteoporosis with porous bone fragility and high risk of bone fracture,Alberto Dell’Era’s Oracle blog CBO: the formula for the …In this post we are going to explore and explain the rationale for the formula used by dbms_stats to [...]

By: Alberto Dell’Era’s Oracle blog » CBO: about the statistical definition of “cardinality” (densities part I)

Sun, 25 Oct 2009 17:44:32 +0000

[...] posts belonging to this series: densities part II densities part III densities part [...]

By: Blogroll Report 09/10/2009-16/10/2009 « Coskan’s Approach to Oracle

Blogroll Report 09/10/2009-16/10/2009 « Coskan’s Approach to Oracle — Wed, 21 Oct 2009 23:29:00 +0000

[...] Alberto Dell’Era-CBO: the formula for the “density” column statistic (densities part II) [...]

By: Alberto Dell’Era’s Oracle blog » CBO: “NewDensity” replaces “density” in 11g, 10.2.0.4 (densities part III)

Fri, 16 Oct 2009 20:33:50 +0000

[...] a previous post, we already discussed the pre-10.2.0.4 scenario: we saw how and when the “density” column statistic [...]

By: Alberto Dell'Era

Alberto Dell'Era — Tue, 13 Oct 2009 08:42:16 +0000

@rudy

In Lewis’ book the formula was introduced using words instead of math symbols but still understandable; Wolfgang uses a more formal mathematical definition.

What I think is “strange” is the assumption about the shape of w(:x); why a customer that has ordered 100 books should ask about the order book list 100 times as frequently as another customer that has ordered only 1 book ? 11g (and 10.2.0.4) corrects this with “NewDensity”, as we will see in the next post.

About E[count(:x)] = “observed count”: that is not true in general, since one might use a statistical model that infer E[count(:x)] from the observed count, for example taking into account the way the table is modified over time (e.g: that an insert_date column is constantly increasing its max value). That would be, of course, very complicated in practice, but statistically sound and possible.

By: Rudy

Rudy — Tue, 13 Oct 2009 08:01:27 +0000

I checked Lewis’ book, but he doesn’t explain how he worked out the density formula – he will publish it in the upcoming book, so we have to wait (how long?!?).

Your explanation about w(:x) is very interesting, although the post is hard for me to read with your notation. I must say it’s convincing, why is it strange for you?

No doubt it’s an easy task for dbms_stats to compute the estimated value for count(:x), or E[count(:x)]… it’s count(:x) indeed, no need to estimate :-)))