Mailing-List: contact user-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@hadoop.apache.org
Received-SPF: neutral (nike.apache.org: local policy)
MIME-Version: 1.0
In-Reply-To: 
 <CABtCYFyGBuP4DQ_Q9bqUVP=zjRvL8qdho5nN0d22bpdoEBAMww@mail.gmail.com>
References: 
 <CABtCYFzx4k36x0-V8M8T_ovSc6sd4EL-Hhn56jNHRbG5UGbQSA@mail.gmail.com>
 <CAND0qzvR7B-5NXNTLwN7mztNotjqM89ZJBKpZ471jStfP9W8Fg@mail.gmail.com>
 <CABtCYFwfD7ZsV88Wq19WLGOfHYEZsFFSx6dZBT__YJGwabjVYQ@mail.gmail.com>
 <CAND0qzsZieXK3Prvun3Ee+nfsD0PZpfgXvd8fzOXz1WZMR=E-A@mail.gmail.com>
 <CABtCYFwVM9HS1-TgGKbpE0wpN9Y8axSk5YYKUsoMqq1RtFig2A@mail.gmail.com>
 <CAND0qzsiWeWkuRdRHOqXt504o48qVKU6akd41MgT6MSrzS4tFg@mail.gmail.com>
 <CABtCYFxts=khibbbjWjkyWmTXV5gkAPB=RpNweB5KOJmOFxKdw@mail.gmail.com>
 <CAND0qztYhQrf9bJ_s=CDerwWZ5iO8CrFz=1AE+oJudY+SJH+-Q@mail.gmail.com>
 <CABtCYFx0Vg53Xp+GVjBHw1S1-RUQiBmoMBt1XAfVUPCkB2DzGg@mail.gmail.com>
 <CAND0qzu6FkD8x8NFgQ4oaYu+93fUztP8HCr=EY7zoaNXLjR4bw@mail.gmail.com>
 <CABtCYFyGBuP4DQ_Q9bqUVP=zjRvL8qdho5nN0d22bpdoEBAMww@mail.gmail.com>
From: Ted Dunning <tdunning@maprtech.com>
Date: Mon, 3 Sep 2012 15:47:01 -0400
Message-ID: 
 <CAND0qzuKWrBu5CR3SsW+vH11+xsPVBjrej5Wir9TzuTZ-aY9uw@mail.gmail.com>
Subject: Re: best way to join?
To: user@hadoop.apache.org
Content-Type: multipart/alternative; boundary=14dae93403f7fd84e804c8d168a1

--14dae93403f7fd84e804c8d168a1
Content-Type: text/plain; charset=ISO-8859-1

On Sun, Sep 2, 2012 at 12:26 PM, dexter morgan <dextermorgan4u@gmail.com>wrote:

> ... Either way, any clustering process requires calculating the distance
> of all points (not between all the points, but of all of them to some
> relative point). Because i'll need a clustering MR job, ill probably use
> it, despite as you said, it has high probability to be correct (not 100%)...
>

This is probably right as stated, but I think that there is confusion here.

Many people assume that each point in the training data has to have
distance computed to all centroids in the clustering.  Even this is not
true.

It is true that you have to compute distance to at least one something, but
not necessarily to all of the clusters.

--14dae93403f7fd84e804c8d168a1
Content-Type: text/html; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable

<br><br><div class=3D"gmail_quote">On Sun, Sep 2, 2012 at 12:26 PM, dexter =
morgan <span dir=3D"ltr">&lt;<a href=3D"mailto:dextermorgan4u@gmail.com" ta=
rget=3D"_blank">dextermorgan4u@gmail.com</a>&gt;</span> wrote:<br><blockquo=
te class=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc so=
lid;padding-left:1ex">

<div dir=3D"ltr"><div>... Either way, any clustering process requires calcu=
lating the distance of all points (not between all the points, but of all o=
f them to some relative point). Because i&#39;ll need a clustering MR job, =
ill probably use it, despite as you said, it has high probability to be cor=
rect (not 100%)...</div>

</div></blockquote><div><br></div><div>This is probably right as stated, bu=
t I think that there is confusion here.</div><div><br></div><div>Many peopl=
e assume that each point in the training data has to have distance computed=
 to all centroids in the clustering. =A0Even this is not true. =A0</div>

<div><br></div><div>It is true that you have to compute distance to at leas=
t one something, but not necessarily to all of the clusters.</div></div>

--14dae93403f7fd84e804c8d168a1--