Hello.

[Please try and set your mail client to send plain text messages.]

Le mer. 26 févr. 2020 à 14:05, CT <chentao...@qq.com> a écrit :
>
> Hi Gilles,
> ------------------&nbsp;Original&nbsp;------------------
> From:&nbsp;"GillesSadowski"<gillese...@gmail.com&gt;;
> Date:&nbsp;Wed, Feb 26, 2020 05:41 PM
> To:&nbsp;"Commons Developers List"<dev@commons.apache.org&gt;;
>
> Subject:&nbsp;Re: [math] Discuss: New feature MiniBatchKMeansClusterer
>
>
>
> &gt;[...]
>
> &gt;&gt; Do you mean this:
> &gt;&gt; &amp;nbsp;* For JIRA issue #MATH-1509 Start a PR with 
> "MiniBatchKMeansClusterer", but without the "ClusterUtils",
> &gt;&gt; despite the duplicate code between "MiniBatchKMeansClusterer" and 
> "KMeansPlusPlusClusterer",
> &gt;&gt; also with "CentroidInitializer" and test code with in a single 
> commit.
> &gt;&gt; &amp;nbsp;* Suggestions like "remove the constructors with default 
> parameters" should apply as a new commit of the PR above,
> &gt;&gt; and tracking by a subtask of JIRA issue #MATH-1509.
> &gt;&gt; &amp;nbsp;* Fire a new JIRA issue for the duplicate code, and start 
> another PR with "ClusterUtils" in,
> &gt;&gt; and extract duplicate code into "ClusterUtils".
> &gt;
> &gt;No, you should start with the smallest possible self-contained PR.
> &gt;For example, why should we commit a code that defines several
> &gt;constructors, while we already know that a second commit should
> &gt;remove them?
> &gt;
> &gt;As you've noticed that some functionality must be factored out of
> &gt;"KMeansPlusPlusClusterer", this should be done first as a separate
> &gt;JIRA issue. IIUC, you propose "ClusterUtils".&nbsp; By reviewing a
> &gt;minimal PR, we should be able to examine whether another
> &gt;approach might be better (than a "utility" class) in order to expose
> &gt;functionality common to all clusterer algorithms.
> &gt;For example, could all "Kmeans" implementations inherit from
> &gt;a common base class?
>
> Do you mean I should fire a JIRA issue about reuse&nbsp;"centroidOf" and 
> "chooseInitialCenters",
> then start a PR and a disscuss about "ClusterUtils"?
> And then&nbsp;start the PR of "MiniBatchKMeansClusterer" after all done?

I cannot guarantee that the whole process will be streamlined.
In effect, you can work on multiple branches (one for each
prospective PR).
I'd say that you should start by describing (here on the ML) the
rationale for "ClusterUtils" (and contrast it with say, a common
base class).
[Only when the design has been agreed on,  a JIRA issue to
implement it should be created in order to track the actual
coding work).]

Gilles

> &gt;[...]

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org
For additional commands, e-mail: dev-h...@commons.apache.org

Reply via email to