On Tue, Dec 6, 2011 at 5:57 AM, Gael Varoquaux
wrote:
> On Mon, Dec 05, 2011 at 11:21:01PM +0100, Andreas Mueller wrote:
>> What you do want is to "transform" the new data so that it
>> is coded using the specified dictionary.
>> I think this is exactly what the sparse encoding method that Olivier
On Mon, Dec 05, 2011 at 01:41:53PM -0500, Alexandre Passos wrote:
> On Mon, Dec 5, 2011 at 13:31, James Bergstra wrote:
> > I should probably not have scared ppl off speaking of a 250-job
> > budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
> > "significant" hyper-parameters,
On Mon, Dec 05, 2011 at 11:21:01PM +0100, Andreas Mueller wrote:
> What you do want is to "transform" the new data so that it
> is coded using the specified dictionary.
> I think this is exactly what the sparse encoding method that Olivier
> referenced is doing.
We would need an intermediate objec
On Mon, Dec 05, 2011 at 10:54:42PM +0100, Olivier Grisel wrote:
> - libsvm uses SMO (a dual solver) and supports non-linear kernels and
> has complexity ~ n_samples^3 hence cannot scale to large n_samples
> (e.g. more than 50k).
> - liblinear uses some kind of fancy coordinate descent (primal or du
On 2011-12-05, at 5:50 PM, Ian Goodfellow wrote:
>
> I think I was mostly confused by the terminology-- I don't consider the code
> to be part of a sparse coding model, nor to be estimated (I am aware that
> sparse coding involves iterative optimization but I don't consider the
> optimizer
> to b
On Mon, Dec 5, 2011 at 5:31 PM, Olivier Grisel wrote:
> 2011/12/5 Andreas Mueller :
>> On 12/05/2011 11:14 PM, Alexandre Gramfort wrote:
I do not understand. I have the dictionary already, so what is being
estimated?
>>> well I am not sure to follow now, but if you have the dictionary t
2011/12/5 Andreas Mueller :
> On 12/05/2011 11:14 PM, Alexandre Gramfort wrote:
>>> I do not understand. I have the dictionary already, so what is being
>>> estimated?
>> well I am not sure to follow now, but if you have the dictionary the
>> only missing part is the coefs of the decomposition.
>>
On 12/05/2011 11:14 PM, Alexandre Gramfort wrote:
>> I do not understand. I have the dictionary already, so what is being
>> estimated?
> well I am not sure to follow now, but if you have the dictionary the
> only missing part is the coefs of the decomposition.
>
> X = dico x coefs
I think there i
> I do not understand. I have the dictionary already, so what is being
> estimated?
well I am not sure to follow now, but if you have the dictionary the
only missing part is the coefs of the decomposition.
X = dico x coefs
Alex
--
2011/12/5 Ian Goodfellow :
> On Mon, Dec 5, 2011 at 4:50 PM, Alexandre Gramfort
> wrote:
>>> One experiment I want to do involves plugging in dictionaries that
>>> were learned with other methods.
>>
>> if you have the dictionaries then use a batch lasso or batch OMP with
>> precomputed gram to ge
On Mon, Dec 5, 2011 at 4:50 PM, Alexandre Gramfort
wrote:
>> One experiment I want to do involves plugging in dictionaries that
>> were learned with other methods.
>
> if you have the dictionaries then use a batch lasso or batch OMP with
> precomputed gram to get the coefficients. That will give y
2011/12/5 Alexandre Passos :
> On Mon, Dec 5, 2011 at 16:26, James Bergstra wrote:
>>
>> This is definitely a good idea. I think randomly sampling is still
>> useful though. It is not hard to get into settings where the grid is
>> in theory very large and the user has a budget that is a tiny fract
2011/12/5 Alexandre Gramfort :
>> One experiment I want to do involves plugging in dictionaries that
>> were learned with other methods.
>
> if you have the dictionaries then use a batch lasso or batch OMP with
> precomputed gram to get the coefficients. That will give you the full
> estimated mode
2011/12/5 Alexandre Gramfort :
> look at
>
> sklearn.multiclass
Indeed, these tools allows the user to build a meta learner with any
multiclass logic on top of a binary classifier implementations (hence
both LinearSVC and SVC can be used as the underlying binary classifier
implementations).
htt
2011/12/5 Ian Goodfellow :
>
> ok, I was using LinearSVC, so I guess I am still not using the dense
> implementation.
>
> Is there a way to use one-against-rest rather than one-against-many
> classification with the SVC class?
What is one-against-many? SVC mutliclass support comes directly from
th
> One experiment I want to do involves plugging in dictionaries that
> were learned with other methods.
if you have the dictionaries then use a batch lasso or batch OMP with
precomputed gram to get the coefficients. That will give you the full
estimated model.
Alex
--
look at
sklearn.multiclass
Alex
On Mon, Dec 5, 2011 at 10:37 PM, Ian Goodfellow
wrote:
> On Mon, Dec 5, 2011 at 4:24 PM, Olivier Grisel
> wrote:
>> 2011/12/5 Ian Goodfellow :
>>> On Fri, Dec 2, 2011 at 3:36 AM, Olivier Grisel
>>> wrote:
2011/12/2 Ian Goodfellow :
> On Fri, Oct 7, 2
I'm interested in doing sparse coding with scikits.learn. It looks
like the way to do this is with
sklearn.decomposition.MiniBatchDictionaryLearning. Am I correct about
that?
If so:
One experiment I want to do involves plugging in dictionaries that
were learned with other methods.
I thought I coul
On Mon, Dec 5, 2011 at 16:26, James Bergstra wrote:
>
> This is definitely a good idea. I think randomly sampling is still
> useful though. It is not hard to get into settings where the grid is
> in theory very large and the user has a budget that is a tiny fraction
> of the full grid.
I'd like t
On Mon, Dec 5, 2011 at 4:24 PM, Olivier Grisel wrote:
> 2011/12/5 Ian Goodfellow :
>> On Fri, Dec 2, 2011 at 3:36 AM, Olivier Grisel
>> wrote:
>>> 2011/12/2 Ian Goodfellow :
On Fri, Oct 7, 2011 at 5:14 AM, Olivier Grisel
wrote:
> 2011/10/7 Ian Goodfellow :
>> Thanks. Yes it d
On Mon, Dec 5, 2011 at 1:41 PM, Alexandre Passos wrote:
> On Mon, Dec 5, 2011 at 13:31, James Bergstra wrote:
>> I should probably not have scared ppl off speaking of a 250-job
>> budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
>> "significant" hyper-parameters, randomly sa
2011/12/5 Ian Goodfellow :
> On Fri, Dec 2, 2011 at 3:36 AM, Olivier Grisel
> wrote:
>> 2011/12/2 Ian Goodfellow :
>>> On Fri, Oct 7, 2011 at 5:14 AM, Olivier Grisel
>>> wrote:
2011/10/7 Ian Goodfellow :
> Thanks. Yes it does appear that liblinear uses only a 64 bit dense format,
>
add it to master
Alex
On Mon, Dec 5, 2011 at 10:16 PM, Satrajit Ghosh wrote:
> hi fabian,
>
> 'ensemble' not in sklearn/setup.py.
>
> config.add_subpackage("ensemble")
> config.add_subpackage("ensemble/tests")
>
> for something like this should i just add it in master? or send a pull
> request.
hi fabian,
'ensemble' not in sklearn/setup.py.
config.add_subpackage("ensemble")
config.add_subpackage("ensemble/tests")
for something like this should i just add it in master? or send a pull
request.
cheers,
satra
--
hello ian,
can you show a snippet of the code you use to train your svm?
and give us the dimensions of your problem?
Alex
On Mon, Dec 5, 2011 at 9:51 PM, Ian Goodfellow wrote:
> On Fri, Dec 2, 2011 at 3:36 AM, Olivier Grisel
> wrote:
>> 2011/12/2 Ian Goodfellow :
>>> On Fri, Oct 7, 2011 at 5:
On Fri, Dec 2, 2011 at 3:36 AM, Olivier Grisel wrote:
> 2011/12/2 Ian Goodfellow :
>> On Fri, Oct 7, 2011 at 5:14 AM, Olivier Grisel
>> wrote:
>>> 2011/10/7 Ian Goodfellow :
Thanks. Yes it does appear that liblinear uses only a 64 bit dense format,
so this memory usage is normal/caused
On Mon, Dec 5, 2011 at 13:44, Olivier Grisel wrote:
> Yes. +1 for a pull request: one could just add a "budget" integer
> argument (None by default) to the existing GridSearchCV class.
Just did that, the pull request is at
https://github.com/scikit-learn/scikit-learn/pull/455
So far no tests. Ho
Dear scikit-learners,
It's about time for a new release. This month of December is rather
busy with the NIPS conference and the coding sprint happening [0] so I
propose to make the release just after holidays, during the first
weeks of January. That should give us enough time to test and
stabilize
On Mon, Dec 5, 2011 at 14:19, Andreas Müller wrote:
> on a related note: what about coarse to fine grid-searches?
> For categorial variables, that doesn't make much sense but
> I think it does for many of the numerical variables.
Coarse-to-fine grid searches (where you expand search in regions ne
On 12/05/2011 07:44 PM, Olivier Grisel wrote:
> 2011/12/5 Alexandre Passos:
>> On Mon, Dec 5, 2011 at 13:31, James Bergstra
>> wrote:
>>> I should probably not have scared ppl off speaking of a 250-job
>>> budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
>>> "significant" hy
2011/12/5 Alexandre Passos :
> On Mon, Dec 5, 2011 at 13:31, James Bergstra wrote:
>> I should probably not have scared ppl off speaking of a 250-job
>> budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
>> "significant" hyper-parameters, randomly sampling around 10-30 points
>
On Mon, Dec 5, 2011 at 13:31, James Bergstra wrote:
> I should probably not have scared ppl off speaking of a 250-job
> budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
> "significant" hyper-parameters, randomly sampling around 10-30 points
> should be pretty reliable.
So pe
I should probably not have scared ppl off speaking of a 250-job
budget. My intuition would be that with 2-8 hyper-parameters, and 1-3
"significant" hyper-parameters, randomly sampling around 10-30 points
should be pretty reliable.
- James
On Mon, Dec 5, 2011 at 1:28 PM, James Bergstra wrote:
>
On Sat, Dec 3, 2011 at 6:32 AM, Olivier Grisel wrote:
>> With regards to the random sampling, I am a bit worried that the results
>> hold for a fair amount of points, and with a small amount of points
>> (which is typically the situation in which many of us hide) it becomes
>> very sensitive to th
If you don't need probabilities you could use
`clf.decision_function(x)` to get the signed distance to the
hyperplane which can also be used as a confidence score.
best,
Peter
2011/12/5 xinfan meng :
> Cool, Thanks!
>
>
> On Mon, Dec 5, 2011 at 8:12 PM, Mathieu Blondel
> wrote:
>>
>> On Mon, De
Cool, Thanks!
On Mon, Dec 5, 2011 at 8:12 PM, Mathieu Blondel wrote:
> On Mon, Dec 5, 2011 at 9:05 PM, xinfan meng wrote:
> > I understand that the LogisticRegression would be similar to LinearSVC in
> > terms of performance. However, I am repeating other person's experiment.
> > Still, Thank y
On Mon, Dec 5, 2011 at 9:05 PM, xinfan meng wrote:
> I understand that the LogisticRegression would be similar to LinearSVC in
> terms of performance. However, I am repeating other person's experiment.
> Still, Thank you.
Paolo Losi has some code that implements Platt's method (internally
used b
I understand that the LogisticRegression would be similar to LinearSVC in
terms of performance. However, I am repeating other person's experiment.
Still, Thank you.
On Mon, Dec 5, 2011 at 8:01 PM, Mathieu Blondel wrote:
> On Mon, Dec 5, 2011 at 8:54 PM, xinfan meng wrote:
> > Hi:
> > I want
On Mon, Dec 5, 2011 at 8:54 PM, xinfan meng wrote:
> Hi:
> I want the classifier to output the class label and its confidence, in
> order to use it in co-training. The predict_proba() in SVC classifier can
> output a confidence. However, this classifier is a bit slow. Can I simulate
> such con
Hi:
I want the classifier to output the class label and its confidence, in
order to use it in co-training. The predict_proba() in SVC classifier can
output a confidence. However, this classifier is a bit slow. Can I simulate
such confidence score (not necessary a probability) with LinearSVM? Th
40 matches
Mail list logo
