def git_init():
# init repo
try:
check_output(['git', 'init'])
except CalledProcessError as err:
error("Error executing 'git init'. Error message:")
print(err.output)
raise SystemExit
# add conf
try:
check_output(['git', 'add', 'abed_conf.py'])
except CalledProcessError as err:
error("Error executing 'git add abed_conf.py'. Error message:")
print(err.output)
raise SystemExit
try:
check_output(['git', 'add', 'abed_tasks.txt'])
except CalledProcessError as err:
error("Error executing 'git add abed_tasks.txt'. Error message:")
print(err.output)
raise SystemExit
try:
check_output(['git', 'add', 'abed_auto.txt'])
except CalledProcessError as err:
error("Error executing 'git add abed_auto.txt'. Error message:")
print(err.output)
raise SystemExit
try:
check_output(['git', 'commit', '-am', '"initialized abed skeleton"'])
except CalledProcessError as err:
error("Error executing initial commit. Error message:")
print(err.output)
raise SystemExit
def run(self):
# this takes over master/worker
if self.task_dict is None:
error("No tasks defined before attempted run. Exiting")
raise SystemExit
mpi_start(self.task_dict)
info("Finished with run command.")
def local(self):
if self.task_dict is None:
error("No tasks defined before attempted run. Exiting")
raise SystemExit
copy_local_files()
mpi_start(self.task_dict, local=True)
info("Finished with run command.")
def __getattr__(self, attr):
if self.__dict__.has_key(attr):
return getattr(self, attr)
else:
error("You probably Britta'd the settings file, "
"I'm missing parameter %s" % attr)
raise SystemExit
def git_add_auto():
try:
check_output(['git', 'add', settings.AUTO_FILE])
except CalledProcessError as err:
error("Error executing 'git add %s'. Error message:" %
settings.AUTO_FILE)
print(err.output)
raise SystemExit
def git_ok():
try:
check_output(['git', 'diff', '--exit-code'])
except CalledProcessError as err:
if err.returncode == 1:
return False
error("Error performing 'git diff --exit-code'. Error message:")
print(err.output)
raise SystemExit
return True
def git_commit_auto():
try:
check_output([
'git', 'commit', '-m', 'automatic commit of auto log file',
settings.AUTO_FILE
])
except CalledProcessError as err:
error("Error performing autocommit for auto log file. Error message:")
print(err.output)
raise SystemExit
info("Automatic auto log file commit")
def git_commit_tbd():
try:
check_output([
'git', 'commit', '-m', 'automatic commit of TBD task file',
settings.TASK_FILE
])
except CalledProcessError as err:
error("Error performing autocommit for TBD file. Error message:")
print(err.output)
raise SystemExit
info("Automatic TBD file commit")
def _unpack_zip(zipfile, all_tasks):
fpath = '%s%s%s' % (settings.ZIP_DIR, os.sep, zipfile)
try:
b = bz2file.BZ2File(fpath)
tar = tarfile.open(fileobj=b)
except tarfile.ReadError:
error("Could not read tarfile: %s" % fpath)
return
mkdir(settings.STAGE_DIR)
tar.extractall(settings.STAGE_DIR)
tar.close()
move_results(all_tasks)
ziplog = settings.ZIP_DIR + os.sep + 'abed_unzipped.txt'
with open(ziplog, 'a') as fid:
fid.write(zipfile + '\n')
def reference_difference(table):
""" Runs Holm's procedure for a reference classifier. """
# Sanity checks
if settings.REFERENCE_METHOD is None:
return None
if (not table.is_summary) or (table.type != AbedTableTypes.RANKS):
return None
if not settings.REFERENCE_METHOD in settings.METHODS:
error("Reference method %s not in list of methods." %
settings.REFERENCE_METHOD)
raise SystemExit
# define constants
N = float(len(settings.DATASETS))
k = float(len(settings.METHODS))
av_ranks = next((row for _id, row in table if _id == 'Average'), None)
av_ranks = [float(x) for x in av_ranks]
ref_idx = settings.METHODS.index(settings.REFERENCE_METHOD)
others = [m for m in settings.METHODS if not m ==
settings.REFERENCE_METHOD]
# Calculate the Z-scores compared to the reference method
Z_scores = [0] * len(others)
P_values = [0] * len(others)
constant = sqrt((6.0*N)/(k*(k+1.0)))
for j, method in enumerate(others):
i = settings.METHODS.index(method)
Z_scores[j] = (av_ranks[ref_idx] - av_ranks[i])*constant
P_values[j] = norm_dist.cdf(Z_scores[j])
# Sort the p-values in ascending order
sorted_pvals = sorted((p, i) for i, p in enumerate(P_values))
# Calculate significant differences following Holm's procedure
significant_differences = [False]*len(others)
CD_threshold = None
for i in range(int(k-1)):
threshold = settings.SIGNIFICANCE_LEVEL/float(k - (i+1))
pval, idx = sorted_pvals[i]
significant_differences[idx] = pval < threshold
if pval > threshold and CD_threshold is None:
CD_threshold = threshold
CD = -1*norm_dist.ppf(CD_threshold) / constant
out = list(zip(others, Z_scores, P_values, significant_differences))
return out, CD
def dataset_completed(dsetfiles, dset, task_dict):
if settings.TYPE == 'ASSESS':
dset_tasks = {
k: v
for k, v in task_dict.iteritems() if v['dataset'] == dset
}
elif settings.TYPE == 'CV_TT':
dset_tasks = {
k: v
for k, v in task_dict.iteritems()
if (v['train_dataset'] == dset[0] and v['test_dataset'] == dset[1])
}
else:
error("Compressing data not supported for TYPE = %s" % settings.TYPE)
raise SystemExit
have_hashes = set([hash_from_filename(f) for f in dsetfiles])
need_hashes = set(dset_tasks.keys())
return have_hashes == need_hashes
def do_work(hsh, task, local=False):
datadir = os.path.join(get_scratchdir(local), 'datasets')
execdir = os.path.join(get_scratchdir(local), 'execs')
if settings.TYPE == 'RAW':
cmd = task.format(datadir=datadir, execdir=execdir)
else:
command = settings.COMMANDS[task['method']]
task['datadir'] = datadir
task['execdir'] = execdir
cmd = command.format(**task)
try:
info("Executing: '%s'" % cmd, color_wrap=False)
output = check_output(cmd, shell=True)
except CalledProcessError as err:
error("There was an error executing: '%s'. Here is the error: %s" %
(cmd, err.output), color_wrap=False)
return
write_output(output, hsh)
info("Finished with %s" % hsh, color_wrap=False)
def copy_worker(local):
comm = MPI.COMM_WORLD
status = MPI.Status()
scratchdir = get_scratchdir(local)
curdir = '%s/releases/current' % settings.REMOTE_DIR
local_results = '%s/results/' % curdir
scratch_results = '%s/results/' % scratchdir
copy_task = ("rsync -avm --include='*.*' -f 'hide,! */' %s %s" %
(scratch_results, local_results))
while True:
if comm.Iprobe(source=0, tag=MPI.ANY_TAG):
comm.recv(obj=None, source=0, tag=MPI.ANY_TAG, status=status)
if status.Get_tag() == KILLTAG:
break
try:
check_output(copy_task, shell=True)
except CalledProcessError:
error("There was an error in the copy task", color_wrap=False)
time.sleep(settings.MW_COPY_SLEEP)
def main():
args = parse_arguments()
skip_init = False
if args.cmd == 'reload_tasks':
skip_init = True
if settings is None:
if not args.cmd == 'init':
error("No ABED configuration file found in this directory. "
"Run 'abed init' to initialize one. Exiting.")
raise SystemExit
skip_init = True
abed = Abed(skip_init=skip_init, skip_cache=args.skip_cache)
info("Running abed command: %s" % args.cmd)
try:
getattr(abed, args.cmd)()
except KeyboardInterrupt:
info("Exiting.")
pass
def compress_dataset(dset):
dsetname = dataset_name(dset)
dsetpath = os.path.join(settings.RESULT_DIR, dsetname)
dsetpath = dsetpath.rstrip(os.sep)
if settings.COMPRESSION == 'bzip2':
extension = 'bz2'
elif settings.COMPRESSION == 'gzip':
extension = 'gz'
elif settings.COMPRESSION == 'lzma':
extension = 'xz'
else:
error("Unknown compression algorithm specified in "
"COMPRESSION configuration. Please check the "
"configuration file.")
raise SystemExit
output_filename = '%s.tar.%s' % (dsetpath, extension)
# lzma will be available in tarfile when abed is ported to Python 3. On
# posix systems we can try compressing with the tar command.
if os.name == 'posix' and settings.COMPRESSION == 'lzma':
try:
cmd = ('XZ_OPT=-9 tar --directory=%s -Jcf %s %s' %
(settings.RESULT_DIR, output_filename, dsetname))
check_output(cmd, stderr=STDOUT, shell=True)
except CalledProcessError:
error("There was an error executing '%s'.")
raise SystemExit
elif settings.COMPRESSION == 'lzma':
error("lzma compression is not yet available for your platform.")
raise SystemExit
else:
mode = 'w:%s' % extension
with tarfile.open(output_filename, mode, compresslevel=9) as tar:
tar.add(dsetpath, arcname=os.path.basename(dsetpath))
def init_config():
cur = os.getcwd()
files = os.listdir(cur)
conf_file = next((x for x in files if x.startswith('abed_conf')), None)
if conf_file is None:
return None
configfile = os.path.realpath(conf_file)
config = {}
try:
exec(open(configfile).read(), config)
except NameError as err:
error("You probably Britta'd the settings file, "
"NameError: %s" % str(err))
except Exception as err:
error("You probably Britta'd the settings file, "
"an error occured parsing it: %s" % str(err))
keys = list(config.keys())
for key in keys:
if not key.upper() == key:
del config[key]
parse_dirs(config, key)
settings = Settings(**config)
return settings
def push(self):
if not git_ok():
error("Git repository has uncommitted changes, not pushing.")
raise SystemExit
info("Starting push")
fab_push()
def check_size():
if not sys.maxsize == 9223372036854775807:
error("Running on a non 64-bit system. This may cause problems with "
"hashes.")
raise SystemExit