when run from a directory containing uncompressed copies of the data from
MIT/LL, run.sh will produce a CSV with one row per test described in the
logs and columns as given in column-names.txt. For example:
iev@bruce brass-unlogger % ls
column-names.txt randomTests run.sh waypoints.py
initialTests readme.md unlogger.py
iev@bruce brass-unlogger % ./run.sh
iev@bruce brass-unlogger % ls
column-names.txt randomTests results.csv unlogger.py
initialTests readme.md run.sh waypoints.py
iev@bruce brass-unlogger %
run.sh is not meant to be especially robust, so it should really only be
used in this way. it's up to you to download and uncompress initialTests
and randomTests and put them in the right place; the .gitignore file
for this repo explicitly bans them from being committed because together
they are about 4.5 GB.
column-names.txt gives the names of the columns in the order that
they'll appear. to move columns around, just move the names in around in
this file. these names can't contain any weird characters --
specifically, if you replace all the `s with_`s, they must be valid
python function names.
unlogger.py reads column-names.txt and calls the functions with the
corresponding names given in file-order.
to add a new column, add a name to column-names.txt in the place you'd
like it to appear, and a function with the same name (modulo underscores
for spaces) in python that defines what its values should be in terms of
the variables in scope in the inner most loop where the call to
locals() is made (those variables will all be in scope inside the
function you define, as a consequence of the meaning of locals())
one caveat is that this means you cannot have any other variables
anywhere in the python with names that are the same as column
headers. for example, there is a column header "final x", so you cannot
have a variable named final_x because it'll shadow the associated
function.
from the tarballs of logs we got, the directory structure of
initialTests and randomTests isn't quite the same; initialTests/*/
looks like randomTests/, with json files that share hashes in their
names with directories. this is reflected in the slightly odd structure
of run.sh.
there is a slight security risk here, in that names in the colums file
are being used to look up into the locals dictionary. this shouldn't be a
problem because this is not world facing: just don't adversarially name
columns things like Robert'); DROP TABLE Students; --.
Once the results.csv file is made, make_maps.py can process it to
create a visualization for each test run. Currently, this means a copy of
the base map with the start, target, final, and obstacle locations marked
for each row in the csv in a PNG named with the test hash number in a
directory called images/. That directory must not exist when you run
make_maps.py. It takes about a half a second to run for each, so this
will be fairly lengthy for the whole results file.
Many thanks to Tom7 for making his font "Hockey is Lif" available for free use on [http://fonts.tom7.com/fonts98.html].