[Topic-models] HLDA NCRP branching off

Jake Sully ccrraappy at gmail.com
Wed Mar 24 00:54:27 EDT 2010


Hi Dave,

Yes, thank you very much for answering my question. Although you did
misread my question initially, I kind of got the answer from your
first reply.

I have another question here. Does all documents have paths of the
same length, where path is the number of nodes between root and leaf
node.

-Jake

On Mon, Mar 22, 2010 at 1:46 AM, David Blei <david.blei at gmail.com> wrote:
> hi jake and all
>
> jon pointed out that i misread your email.  (thanks jon!)  by "second
> down" i meant "off the root and on down."  it's not as though the root
> has only one child.  it can have many children.  i hope this is clear.
>
> best
> dave
>
> On Sun, Mar 21, 2010 at 9:23 AM, David Blei <david.blei at gmail.com> wrote:
>> hi jake
>>
>> our goal was to infer a single tree that describes the collection.
>> that is why the root node is fixed.  an alternative model, and one
>> that is totally reasonable, is one where there are multiple root
>> nodes.
>>
>> best
>> dave
>>
>> On Sat, Mar 20, 2010 at 12:49 AM, Jake Sully <ccrraappy at gmail.com> wrote:
>>> Hi List,
>>>
>>> I have some questions regarding the branching of NCRP with reference
>>> to this earlier post by David Blei,
>>>
>>> https://lists.cs.princeton.edu/pipermail/topic-models/2006-October/000031.html
>>>
>>> quote:
>>> what are the possible paths that can be drawn?  first, there is one
>>> for each existing leaf in the tree.  then, there is a new one for
>>> each level---a new branch can be spawned at any level from the second
>>> down.  in total, this is the number of nodes in the currently
>>> populated tree.  for a particular path c, when one of the levels is a
>>> new restaurant, the quantity (C) above is 0.
>>>
>>> I want to ask regarding this sentence "a new branch can be spawned at
>>> any level from the second down"
>>>
>>> May I know why is it second down and not first down?
>>> That means we can start spawning new nodes off the root node.
>>>
>>> -Jake
>>> _______________________________________________
>>> Topic-models mailing list
>>> Topic-models at lists.cs.princeton.edu
>>> https://lists.cs.princeton.edu/mailman/listinfo/topic-models
>>>
>>
>


More information about the Topic-models mailing list