[SciPy-user] Presentation of pymachine, a python package for machine learning

[David Cournapeau]

Gael Varoquaux wrote:
> On Mon, May 14, 2007 at 09:13:34AM +0200, Matthieu Brucher wrote:
>>    My view was still a GUI framework, like Orange, but with more flexibility.
>>    For instance, every scipy module or scikit could have a wrapper in this
>>    framework and be used efficiently.
>
> I am not to sure what you call machine learning, but the GUI of Orange
> reminds me of LabView.
When I am cynical, I say that machine learning is a cute name for 
maximization over constraints. More seriously, machine learning is more 
or less synonymous with AI (at least a pretty big subfield of AI), and 
tries to find methods for finding pattern in data.

Concrete examples:
    - you have a dataset of music extracts, and want to find the 
extracts similar to a given melody. So you want to find data "similar" 
to your extract, according to some "similarity" measures.
    - you have a dataset of brain images with some diseases, and you 
want to find the disease corresponding to a new patient.
>
> I was thinking of how to do a labview-like UI with Python... I think
> Traits[1] would be a great solution.
One of the thing I am not sure about is the choice of a toolkit for the 
GUI.
    - matplotlib is great for some widget which are pretty complete 
(spectrogram, etc...), but not really usable for general GUI.
    - TraitsUI looks great, but I am a bit worried that it may be 
difficult to install on some platforms.
    - Other toolkits: problem similar to TraitsUI. Except Tkinter, which 
I really don't intend to use, none of the GUI toolkit for python are 
standards.

    Before choosing a toolkit, I will have to work a bit with TraitsUI. 
Do you know if it is easy to add matplotlib widgets inside TraitsUI ? I 
am always a bit concerned with toolkit choices, because on Unix, you 
cannot easily mix different toolkits inside one process, and this can 
makes things really hairy if you want to use widgets from different 
toolkits.

David