Discussion:
More module renaming
(too old to reply)
Gael Varoquaux
2010-11-29 14:39:11 UTC
Permalink
Hi all,

Two suggestions in this mail (from crazy, to crazier):

* While we are talking of module renaming, and as this is going to teribly
annoy our users, I was thinking that I would prefer:

feature_extraction -> feature_extract
feature_selection -> feature_select

The reason is two-fold: it would be more consistent with the rest of the
scikit and scipy (as 'cluster' and not 'clustering'), and it is exactly 3
characters shorter (think of the keyboard!).

But maybe I am really nickpicking and waisting my time (I should be
working on my deadline,... ooops, my boss reads this mailing list). What
do others think.

* Finally (let's go crazy about API changes), I would like to have a
sub-package for model selection, because we have a lot of code related to
this that is scattered a bit. This brings a few questions:

1. Is it a good idea at all?
2. How should it be named (model_select seems long and heavy, maybe
'tune', but that seems a bit 'Jacky', and not very scientific)?
3. What should go in there. I was thinking of:
- metrics
- cross_val
- grid_search
In the long run, I see also the option of having other wrapping
cross-validation estimators, as the GridSearchCV that would implement
other strategies that grid search.

Of course, all the objects useful for the end-user would be imported in
the __init__ of this sub-package, so that the user wouldn't have to go 2
levels deep.

Feedback?

Gaël
Olivier Grisel
2010-11-29 14:50:38 UTC
Permalink
Post by Gael Varoquaux
Hi all,
* While we are talking of module renaming, and as this is going to teribly
   feature_extraction -> feature_extract
   feature_selection  -> feature_select
The reason is two-fold: it would be more consistent with the rest of the
scikit and scipy (as 'cluster' and not 'clustering'), and it is exactly 3
characters shorter (think of the keyboard!).
I am +0 for this (the macbook pro keyboards are pretty robust anyway).
Post by Gael Varoquaux
But maybe I am really nickpicking and waisting my time (I should be
working on my deadline,... ooops, my boss reads this mailing list). What
do others think.
:)
Post by Gael Varoquaux
* Finally (let's go crazy about API changes), I would like to have a
sub-package for model selection, because we have a lot of code related to
   1. Is it a good idea at all?
   2. How should it be named (model_select seems long and heavy, maybe
      'tune', but that seems a bit 'Jacky', and not very scientific)?
       - metrics
       - cross_val
       - grid_search
     In the long run, I see also the option of having other wrapping
     cross-validation estimators, as the GridSearchCV that would implement
     other strategies that grid search.
Of course, all the objects useful for the end-user would be imported in
the __init__ of this sub-package, so that the user wouldn't have to go 2
levels deep.
I am -0 on this one. I like flat package names better but not strongly
opposed to the model_select package (but I do not like the "tune" name
since it's not part of the usual machine learning terminology for
model selection).
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Mathieu Blondel
2010-11-29 15:02:39 UTC
Permalink
On Mon, Nov 29, 2010 at 11:39 PM, Gael Varoquaux
Post by Gael Varoquaux
* While we are talking of module renaming, and as this is going to teribly
   feature_extraction -> feature_extract
   feature_selection  -> feature_select
I don't see how it is more consistent with the rest of the scikit.
Either you see "extract" and "select" as abbreviations and we have
decided to go for full intelligible names. Or you see them as verbs
and this is not consistent with the rest either. So -1 for me.
Post by Gael Varoquaux
The reason is two-fold: it would be more consistent with the rest of the
scikit and scipy (as 'cluster' and not 'clustering'), and it is exactly 3
characters shorter (think of the keyboard!).
My primary feeling when I read "cluster" is to read it as a noun,
rather than a verb.
Post by Gael Varoquaux
   2. How should it be named (model_select seems long and heavy, maybe
      'tune', but that seems a bit 'Jacky', and not very scientific)?
Thank you for giving me a good laugh with the use of "Jacky" :)
Post by Gael Varoquaux
       - metrics
       - cross_val
       - grid_search
     In the long run, I see also the option of having other wrapping
     cross-validation estimators, as the GridSearchCV that would implement
     other strategies that grid search.
I don't dislike the idea although I think core modules like that (by
core I mean which can benefit to all algorithms) are good candidates
to be at the root of the scikit.

# @Gael and others: can you give your opinion on stochastic_gradient
vs stochastic in the other thread so that Peter can take his decision.
Thanks!

Mathieu
Gael Varoquaux
2010-11-29 15:09:45 UTC
Permalink
Post by Mathieu Blondel
vs stochastic in the other thread so that Peter can take his decision.
IMHO it boils down to whether we will have other stochastic algorithms. I
can't answer that question, I am too illiterate in stochastic algorithms.

It might be nice, though to have similar sub-structure in the various
sub-packages: for instance it seems that sparse will appear in many
subpackages. I can see room for stochastic in many subpackages too.

I guess it's a +0 from me, and I'd rather have the wise give their
opinion.

G
Peter Prettenhofer
2010-11-29 16:13:11 UTC
Permalink
I'd prefer to name the package `stochastic_gradient` since this is
more specific; consider the stochastic coordinate descent algorithm
that Alexandre (Passos) talked about - should it go into the
stochastic (sub-)package or in the coordinate descent (sub-)package?
With `stochastic_gradient` we don't run into such problems.

I think the more important question is whether or not
`stochastic_gradient` should be placed in `linear_model`. Currently, I
think this is the better decision but maybe at some point in the
future we will have multi-layer perceptrons trained via SGD and we
have to revise that decision...

best,
Peter
Post by Gael Varoquaux
Post by Mathieu Blondel
vs stochastic in the other thread so that Peter can take his decision.
IMHO it boils down to whether we will have other stochastic algorithms. I
can't answer that question, I am too illiterate in stochastic algorithms.
It might be nice, though to have similar sub-structure in the various
sub-packages: for instance it seems that sparse will appear in many
subpackages. I can see room for stochastic in many subpackages too.
I guess it's a +0 from me, and I'd rather have the wise give their
opinion.
G
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
Gael Varoquaux
2010-11-29 16:31:18 UTC
Permalink
Post by Peter Prettenhofer
I'd prefer to name the package `stochastic_gradient` since this is
more specific; consider the stochastic coordinate descent algorithm
that Alexandre (Passos) talked about - should it go into the
stochastic (sub-)package or in the coordinate descent (sub-)package?
With `stochastic_gradient` we don't run into such problems.
I would have put it in a generic 'stochastic', because I sympathize with
the dumb user who does not know precisely where to look for what he
wants. However the other view is also consistent.
Post by Peter Prettenhofer
I think the more important question is whether or not
`stochastic_gradient` should be placed in `linear_model`. Currently, I
think this is the better decision but maybe at some point in the
future we will have multi-layer perceptrons trained via SGD and we
have to revise that decision...
For both of these questions I believe that I am not the right person to
give an answer. Probably you and Alexandre (Passos) are the persons who
are going to be putting the most effort in these parts of the code, so
you should feel comfortable with the decision.

Gaël
Olivier Grisel
2010-11-29 16:52:39 UTC
Permalink
Post by Peter Prettenhofer
I'd prefer to name the package `stochastic_gradient` since this is
more specific; consider the stochastic coordinate descent algorithm
that Alexandre (Passos) talked about - should it go into the
stochastic (sub-)package or in the coordinate descent (sub-)package?
With `stochastic_gradient` we don't run into such problems.
+1
Post by Peter Prettenhofer
I think the more important question is whether or not
`stochastic_gradient` should be placed in `linear_model`. Currently, I
think this is the better decision but maybe at some point in the
future we will have multi-layer perceptrons trained via SGD and we
have to revise that decision...
If we are to implement multilayer perceptrons with logistic activation
functions, we will probably have an array of predictors as output
instead of working with one predictor at a time and combining them
with OVA.

Hence the algorithmic structure of the code will be very different
from the existing code base and we might have very few code in common
and might be better implemented in separate independent packages
anyway.

So I don't mind if we move the current sgd package as
linear_model.stochastic_gradient.
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Peter Prettenhofer
2010-11-29 17:04:34 UTC
Permalink
OK - I think we have a consensus.

To sum up, I'll:
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)

thanks,
Peter

PS: If you have concerns just send me an email - I'll not push
anything to the main repo until Fabian agrees.
Post by Olivier Grisel
Post by Peter Prettenhofer
I'd prefer to name the package `stochastic_gradient` since this is
more specific; consider the stochastic coordinate descent algorithm
that Alexandre (Passos) talked about - should it go into the
stochastic (sub-)package or in the coordinate descent (sub-)package?
With `stochastic_gradient` we don't run into such problems.
+1
Post by Peter Prettenhofer
I think the more important question is whether or not
`stochastic_gradient` should be placed in `linear_model`. Currently, I
think this is the better decision but maybe at some point in the
future we will have multi-layer perceptrons trained via SGD and we
have to revise that decision...
If we are to implement multilayer perceptrons with logistic activation
functions, we will probably have an array of predictors as output
instead of working with one predictor at a time and combining them
with OVA.
Hence the algorithmic structure of the code will be very different
from the existing code base and we might have very few code in common
and might be better implemented in separate independent packages
anyway.
So I don't mind if we move the current sgd package as
linear_model.stochastic_gradient.
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
Olivier Grisel
2010-11-29 17:11:40 UTC
Permalink
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Go! Don't forget to update the examples and the docstrings.

$ grep -ri sgd *
$ grep -ri RegressorSGD *
$ grep -ri ClassifierSGD *

$ make clean inplace test test-doc
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Yaroslav Halchenko
2010-11-29 17:46:52 UTC
Permalink
Hi Everyone,

not to distract you from this thread, but rather to go "crazy" with API changes
as Gael wanted ;-) , you might find following bash function of some use
for your refactorings:

git-sedi () {
git grep -l $1 | xargs sed -i -e "s/$1/$2/g"
}

so just do

git sedi ClassifierSGD SGDClassifier

and you are done

and of cause, "git grep" is your friend here, so you grep only in the materials
under VCS

Cheers,
Post by Olivier Grisel
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Go! Don't forget to update the examples and the docstrings.
$ grep -ri sgd *
$ grep -ri RegressorSGD *
$ grep -ri ClassifierSGD *
$ make clean inplace test test-doc
--
=------------------------------------------------------------------=
Keep in touch www.onerussian.com
Yaroslav Halchenko www.ohloh.net/accounts/yarikoptic
Olivier Grisel
2010-11-29 17:48:56 UTC
Permalink
Post by Yaroslav Halchenko
Hi Everyone,
not to distract you from this thread, but rather to go "crazy" with API changes
as Gael wanted ;-) , you might find following bash function of some use
git-sedi () {
   git grep -l $1 | xargs sed -i -e "s/$1/$2/g"
}
so just do
git sedi ClassifierSGD SGDClassifier
and you are done
and of cause, "git grep" is your friend here, so you grep only in the materials
under VCS
Thanks for the tip!
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Peter Prettenhofer
2010-11-29 18:00:42 UTC
Permalink
Thanks!
Post by Yaroslav Halchenko
Hi Everyone,
not to distract you from this thread, but rather to go "crazy" with API changes
as Gael wanted ;-) , you might find following bash function of some use
git-sedi () {
   git grep -l $1 | xargs sed -i -e "s/$1/$2/g"
}
so just do
git sedi ClassifierSGD SGDClassifier
and you are done
and of cause, "git grep" is your friend here, so you grep only in the materials
under VCS
Cheers,
Post by Olivier Grisel
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Go! Don't forget to update the examples and the docstrings.
$ grep -ri sgd *
$ grep -ri RegressorSGD *
$ grep -ri ClassifierSGD *
$ make clean inplace test test-doc
--
=------------------------------------------------------------------=
Keep in touch                                     www.onerussian.com
Yaroslav Halchenko                 www.ohloh.net/accounts/yarikoptic
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
Gael Varoquaux
2010-11-29 18:12:48 UTC
Permalink
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Thanks Peter for getting the discussion going and driving it to a
conclusion!

Gaël
Peter Prettenhofer
2010-11-29 17:09:29 UTC
Permalink
Post by Olivier Grisel
[..]
If we are to implement multilayer perceptrons with logistic activation
functions, we will probably have an array of predictors as output
instead of working with one predictor at a time and combining them
with OVA.
Hence the algorithmic structure of the code will be very different
from the existing code base and we might have very few code in common
and might be better implemented in separate independent packages
anyway.
certainly, at least they will share the same optimization technique...
but structuring the models according to the underlying optimization
procedure is probably not the best decision from a users' point of
view... so forget what I just said :-)

best,
Peter
--
Peter Prettenhofer
Fabian Pedregosa
2010-11-30 07:48:19 UTC
Permalink
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)

Fabian
Peter Prettenhofer
2010-11-30 10:39:49 UTC
Permalink
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?

best,
Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
Peter Prettenhofer
2010-11-30 11:24:12 UTC
Permalink
Hi everybody,

I issued a pull request for the SGD module renaming.

I haven't restructured the docs and examples yet (they are updated
though) - I think we should clarify this first: should I move
examples/sgd into examples/linear_model and rename it again to
stochastic_gradient? (I think we should definitely do that)

What about doc/modules/sgd.rst - it is rather long - should I shorten
it a bit and add it as a subsection to linear_model.rst or should we
create a "sub-page"?

best,
Peter
Post by Peter Prettenhofer
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?
best,
 Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
--
Peter Prettenhofer
Olivier Grisel
2010-11-30 12:44:17 UTC
Permalink
Post by Peter Prettenhofer
Hi everybody,
I issued a pull request for the SGD module renaming.
I haven't restructured the docs and examples yet (they are updated
though) - I think we should clarify this first: should I move
examples/sgd into examples/linear_model and rename it again to
stochastic_gradient? (I think we should definitely do that)
+1
Post by Peter Prettenhofer
What about doc/modules/sgd.rst - it is rather long - should I shorten
it a bit and add it as a subsection to linear_model.rst or should we
create a "sub-page"?
Maybe you can write an executive overview as a paragraph in the
linear_model.rst page that gives comparison elements with CD and LARS
for regression and LinearSVC for classification and link to the
detailed SGD documentation as a sub-page.
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Gael Varoquaux
2010-11-30 15:24:04 UTC
Permalink
----- Original message -----
Post by Olivier Grisel
Maybe you can write an executive overview as a paragraph in the
linear_model.rst page that gives comparison elements with CD and LARS
for regression and LinearSVC for classification and link to the
detailed SGD documentation as a sub-page.
+1

It would be great to have a comparison table between the methods at the beginning of the linear model page.
Gael Varoquaux
2010-11-30 15:21:46 UTC
Permalink
I am at a phd defense, on my mobile phone, so I can't give a long answer, but my gut feeling is (on part of your question): don't reduce the docs, they are good. Maybe effort can be put in finding the right order, the right prioritization, of the order.

Gael

----- Original message -----
Post by Peter Prettenhofer
Hi everybody,
I issued a pull request for the SGD module renaming.
I haven't restructured the docs and examples yet (they are updated
though) - I think we should clarify this first: should I move
examples/sgd into examples/linear_model and rename it again to
stochastic_gradient? (I think we should definitely do that)
What about doc/modules/sgd.rst - it is rather long - should I shorten
it a bit and add it as a subsection to linear_model.rst or should we
create a "sub-page"?
best,
  Peter
Post by Peter Prettenhofer
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?
best,
 Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game
by optimizing for Intel(R) Graphics Technology. Get started today
with the Intel(R) Software Partner Program. Five $500 cash prizes
are up for grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
--
Peter Prettenhofer
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Gael Varoquaux
2010-11-30 15:26:33 UTC
Permalink
IMHO they should be imported in the __init__ of the linear_model sub-package.

Cheers,

Gael

----- Original message -----
Post by Peter Prettenhofer
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?
best,
  Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Fabian Pedregosa
2010-12-01 10:55:55 UTC
Permalink
To keep everyone posted: discussion continues on the pull request:

https://github.com/scikit-learn/scikit-learn/pull/25

On Tue, Nov 30, 2010 at 4:22 PM, Gael Varoquaux
Post by Gael Varoquaux
IMHO they should be imported in the __init__ of the linear_model sub-package.
Cheers,
Gael
----- Original message -----
Post by Peter Prettenhofer
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?
best,
   Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
Peter Prettenhofer
2010-12-02 10:20:03 UTC
Permalink
Here goes the next pull request (for examples + docs)
https://github.com/scikit-learn/scikit-learn/pull/26

best,
Peter
Post by Fabian Pedregosa
https://github.com/scikit-learn/scikit-learn/pull/25
On Tue, Nov 30, 2010 at 4:22 PM, Gael Varoquaux
Post by Gael Varoquaux
IMHO they should be imported in the __init__ of the linear_model sub-package.
Cheers,
Gael
----- Original message -----
Post by Peter Prettenhofer
BTW: should I add SGDClassifier and SGDRegressor to the linear_model
namespace or just stochastic_gradient?
best,
   Peter
Post by Fabian Pedregosa
On Mon, Nov 29, 2010 at 6:04 PM, Peter Prettenhofer
Post by Peter Prettenhofer
OK - I think we have a consensus.
1. put `sgd` into `linear_model`
2. rename `sgd` to `stochastic_gradient`
3. rename ClassifierSGD to SGDClassifier (same for RegressorSGD)
Perfecto :-)
Fabian
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for
grabs. http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
------------------------------------------------------------------------------
Increase Visibility of Your 3D Game App & Earn a Chance To Win $500!
Tap into the largest installed PC base & get more eyes on your game by
optimizing for Intel(R) Graphics Technology. Get started today with the
Intel(R) Software Partner Program. Five $500 cash prizes are up for grabs.
http://p.sf.net/sfu/intelisp-dev2dev
_______________________________________________
Scikit-learn-general mailing list
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
--
Peter Prettenhofer
Olivier Grisel
2010-11-29 15:09:35 UTC
Permalink
Post by Mathieu Blondel
vs stochastic in the other thread so that Peter can take his decision.
Thanks!
I am ok with both stochastic_gradient and stochastic although I would
prefer stochastic_gradient.

As for scikits.learn.linear_model vs scikits.learn.linear what is your
preference?
--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel
Gael Varoquaux
2010-11-29 15:10:48 UTC
Permalink
Post by Olivier Grisel
As for scikits.learn.linear_model vs scikits.learn.linear what is your
preference?
+0 for linear_model

G
Continue reading on narkive:
Loading...