[SciPy-User] Trying hand-writing recognition with scikits.learn
Klonuo Umom
klonuo at gmail.com
Sun May 8 02:11:16 EDT 2011
Re-reading what I wrote yesterday may seem un-balanced, but please
consider that it is not with intend to be disrespectful toward any work
been made. Also consider that English is not my native language and I
can't always describe things the way I wanted.
> For me, achieving things like that with Gnuplot would be incredibly
> frustrating. The gnuplot learning curve is steep :)
>
> For your feedback to be more constructive, it might be good if you also
> provided suggestions on how to improve the situation.
>
> What do you want to have? Did you read the existing beginner-level
> documentation of Matplotlib (i.e. User's Guide)? If yes, was it useful
> or not? What in the specific example you cite was difficult to do?
It is useful. I have it together with PACKT's 'Matplotlib for Python
Developers'. Both are useful and web page is just as beautiful as
useful, showing state of the art structuring - I can find anything I'm
interested in, in 2 clicks. And I feel terrible complaining about it,
but that's what happened.
I don't have example data right here, but it was data from 30
participants divided in 5 bins. I looked at 'table_demo.py' from gallery
page and 'bars stacked' graph from Chapter 9 in PACKT book. IIRC I
could not make bars to be presented at 100% each (so all at same overall
level) instead as appended absolute values, as shown in both examples I
followed. It sounds trivial but I had problem with 'yoff' IIRC and
getting to know mechanism behind stacked bars, then just left it as I
had to make those plots in time.
I have gnuplot since I know I made my first plot, and that
could suggest I need to spend more time with matplotlib to gain
advantage of using advanced plotting library in python code. And that's
fine as I plan to do that, and I used what I experienced as a prolog to
my scikits.learn problem.
> The best place to start would be to read the tutorial part of
> the user guide
>
> http://scikit-learn.sourceforge.net/tutorial.html
>
> It seems to explain exactly the example you mention in detail.
Been there. First I thought, that this is too advanced for me and I
never liked statistics really, but when I saw 'Recognizing hand-written
digits' I got interested for some reason and went on trying,
nonetheless. If I open the csv table I see 64 column and many rows
presented by 0-16 digits. Each row presents image. If I follow link for
the source from which this table is distilled, it seems it's some data
based on output from 'WACOM PL-100V pressure sensitive tablet with an
integrated LCD display and a cordless stylus'. I don't see connection
between 'http://archive.ics.uci.edu/ml/machine-learning-databases/pendigits/'
and '..\scikits\learn\datasets\data\digits.csv.gz'. Or I don't see how
can I make my test model.
Cheers
More information about the SciPy-User
mailing list