[ML] [New Feature] Trainers as pipeline parameters that can be varied

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

[ML] [New Feature] Trainers as pipeline parameters that can be varied

dmitrievanthony
Hi all,A machine learning pipeline implemented in  IGNITE-9158
<https://issues.apache.org/jira/browse/IGNITE-9158>   (see discussion  here
<http://apache-ignite-developers.2346864.n4.nabble.com/ML-Machine-Learning-Pipeline-Improvement-tt32772.html>
) supports hyperparameters variation, but not trainers variation so far.Our
reference-framework scikit-learn (according to  documentation
<http://scikit-learn.org/stable/modules/pipeline.html#pipeline>  ) allows to
variate trainers and preprocessors the following way:>>> param_grid =
dict(reduce_dim=[None, PCA(5),
PCA(10)],...&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;clf=[SVC(),
LogisticRegression()],...&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;&emsp;clf__C=[0.1,
10, 100])>>> grid_search = GridSearchCV(pipe, param_grid=param_grid)I think
it would be a great improvement for our ML pipeline.Alexey Zinoviev, it
would be awesome if you as an author of original ML pipeline take a look at
this proposal.



--
Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/
Reply | Threaded
Open this post in threaded view
|

Re: [ML] [New Feature] Trainers as pipeline parameters that can be varied

dmitrievanthony
In case HTML doesn't work I'll duplicate the message.

Hi all,

A machine learning pipeline implemented in
https://issues.apache.org/jira/browse/IGNITE-9158 (see discussion  here
http://apache-ignite-developers.2346864.n4.nabble.com/ML-Machine-Learning-Pipeline-Improvement-tt32772.html)
supports hyperparameters variation, but not trainers variation so far.

Our  reference-framework scikit-learn (according to  documentation
http://scikit-learn.org/stable/modules/pipeline.html#pipeline) allows to
variate trainers and preprocessors the following way:

>>> param_grid = dict(reduce_dim=[None, PCA(5), PCA(10)],
...                   clf=[SVC(), LogisticRegression()],
...                   clf__C=[0.1, 10, 100])
>>> grid_search = GridSearchCV(pipe, param_grid=param_grid)

I think it would be a great improvement for our ML pipeline.

Alexey Zinoviev, it would be awesome if you as an author of original ML
pipeline take a look at this proposal.



--
Sent from: http://apache-ignite-developers.2346864.n4.nabble.com/
Reply | Threaded
Open this post in threaded view
|

Re: [ML] [New Feature] Trainers as pipeline parameters that can be varied

Alexey Zinoviev