[WIP] K-means

[jbarnoud]

To do:

• convert the old API calls to the new ones
• write unit tests
• write the command line interface
• document the API
• document the CLI

[pierrepo]

Dear @jbarnoud and @HubLot can you please complete this PR?

[HubLot]

I can manage the integration of the k-means into PBxplore (and documentation).
But before that, I have to know if the k-means implementation is correct and finished.

[jbarnoud]

I'll do it. I got stuck at testing if the implementation is correct as I
do not really have a reference to compare with.

On 08-04-16 13:58, Hub wrote:

I can manage the integration of the k-means into PBxplore (and
documentation).
But before that, I have to know if the k-means implementation is
correct and finished.

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#106 (comment)

[alexdb27]

simulated data do not work ?

[jbarnoud]

I got stuck at finding simulated data that were not too artificial to be
relevant. I'll have a new look at this this week-end.

On 08-04-16 14:27, Alexandre G. de Brevern wrote:

simulated data do not work ?

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#106 (comment)

[alexdb27]

you have not simulated data ?

[jbarnoud]

I did some tests. I'll have to go back to it.

On 08-04-16 14:33, Alexandre G. de Brevern wrote:

you have not simulated data ?

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#106 (comment)

[pierrepo]

Maybe if your implementation gives similar results to what is obtained with scikit-learn (http://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html) it should be ok?

[jbarnoud]

On 08/04/16 15:14, Pierre Poulain wrote:

Maybe if your implementation gives similar results to what is obtained
with scikit-learn
(http://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html)
it should be ok?

—
You are receiving this because you were mentioned.
Reply to this email directly or view it on GitHub
#106 (comment)

No because I used completely different definitions for the centers and
the distances.

[pierrepo]

Hi @jbarnoud Could you have a look to this WIP?

[alexdb27]

hello, how is it ?

[pierrepo]

After discussion with @jbarnoud We probably keep this feature for another release of PBxplore.
@jbarnoud is it possible to remove all trace of clustering from the master branch?

[jbarnoud]

On 27/06/16 18:56, Pierre Poulain wrote:

After discussion with @jbarnoud https://github.com/jbarnoud We
probably keep this feature for another release of PBxplore.
@jbarnoud https://github.com/jbarnoud is it possible to remove all
trace of clustering from the master branch?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#106 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/ABUWuqBKnoyEhDSiLmJH6kl6hagYBJ8jks5qQACqgaJpZM4GtFZ1.

It is in my TODO list for tomorrow. In a miraculous universe where I
manage to follow my plans, there should be a PR very soon.

By the way, this PR is a working tool. You can play with it if PBkmeans
if you want to have fun.

[alexdb27]

Ok
Le 27 juin 2016 18:56, "Pierre Poulain" [email protected] a écrit :

After discussion with @jbarnoud https://github.com/jbarnoud We probably
keep this feature for another release of PBxplore.
@jbarnoud https://github.com/jbarnoud is it possible to remove all
trace of clustering from the master branch?

—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
#106 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/AJWGcGKeufrGobRWYyrOEAZjBTqpafq3ks5qQACqgaJpZM4GtFZ1
.

[alexdb27]

PR for what

[jbarnoud]

This PR (#106) implements PBkmeans.

I have a branch somewhere that makes hclust work with python only. But ward is not available because of an issue in scipy.

I will submit a PR to remove all traces of clustering in PBxplore. It is not very difficult to have something that will somehow group sequences, but I would like to assess the strength and drawbacks of the different algorithms rather than just releasing an algo at random.

[pierrepo]

@jbarnoud can I add this branch to Pbxplore repo?

[pierrepo]

@jbarnoud can I add this branch to Pbxplore repo?

[jbarnoud]

On 04/07/16 15:18, Pierre Poulain wrote:

@jbarnoud https://github.com/jbarnoud can I add this branch to
Pbxplore repo?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#106 (comment),
or mute the thread
https://github.com/notifications/unsubscribe/ABUWujqbww_TupmBiXBjPfLAphtZPRhzks5qSQgygaJpZM4GtFZ1.

Nobody reviewed it. Neither the code nor the science behind it.

[pierrepo]

@jbarnoud but as a "dev_pbkmeans" branch? So that you can clean/remove the clustering from the master branch.

[jbarnoud]

Why not.
Le 4 juil. 2016 3:46 PM, Pierre Poulain [email protected] a écrit :@jbarnoud but as a "dev_pbkmeans" branch? So that you can clean/remove the clustering from the master branch.

—You are receiving this because you were mentioned.Reply to this email directly, view it on GitHub, or mute the thread.

[pierrepo]

Hi @jbarnoud
I changed the target branch for this PR.
Could you refresh the PR?