[Databionic-esom-devel] About logging, cvs, data safety and floodfill

[Christian S. <st...@Ma...>]

Hi everyone,

the first version of this message was catched by the fourceforge
spamfilter, because the original title 'esom news' sounded to much like
'some stock news' :-) I hope this one will reach you.

1) logging
2) data safety
3) floodfill

1) logging
-----------
I'm preparing a large, but trivial commit and I want to avoid any cvs
collisions. I'd therefore like to know who of you is working on esom-code
at the moment to avoid cvs collisions.

The point is to replace all log.debug(".."); with
if(log.isDebugEnabled())log.debug("..");. That is because when
log.debug("foo" + intValue + "bar" + object.getValue()); is called, the
parameter is concatenated before the logger has a chance to decide wether
it is needed or not and constructing this string may be arbitrarily
complex.

I ran ten trainings of hepta.lrn before and ten after the changes with
standard settings and under the same conditions. With a mean of around 35
seconds there was a speedup of around 0,5 seconds afterwards. Of course
also startup, rendering and anything else will profit.

I'm considering to apply the same to all log levels as one might want to
switch logging off, when e.g. processing larger batch-jobs. Are there any
objections to this?

2) data safety
---------------
I extended JFileChooser to check for existing files. The User is asked for
confirmation now if he is about to overwrite a file. This might not apply
to the slideshow and does definitely not apply for training, which will
still overwrite existing files. I have no reason to believe, that it won't
work under windows or mac os, but I did not check.
For Training there has to be thought of something.

3) floodfill
-------------
I Fixed the add and substract behaviour for floodfill. It's now possible
to classify only the walls of a map with a few simple clicks. E.g. for
hepta, do the following:
- train a wts
- change to classmask tab
- activate floodfill, set treshold to one and click on map. that will
result in the whole map being covered by one class.
- set treshold to 0.4 and select the class resulting from the previous
step in the table.
- hold shift while clicking once in each of the seven clusters.
- now only the walls are classified

regards,
Christian