def get_benchmark_list_by_name(database_name):
with database.connect(db_name=database_name) as session:
return [
r.name for r in session.execute(
'SELECT DISTINCT name from benchmarks ORDER BY benchmark_id DESC'
)
]
def resume_benchmark(benchmark_id, nstruct=None):
qsub_command = 'qsub',
benchmark_command = 'loop_benchmark.py', benchmark_id
# You get weird errors if you forget to cast nstruct from string to int.
if nstruct is not None: nstruct = int(nstruct)
# Read the job parameters from the database.
with database.connect() as session:
benchmark = session.query(database.Benchmarks).get(benchmark_id)
num_pdbs = len(benchmark.input_pdbs)
# Make sure the right version of rosetta is being used.
git_commit = subprocess.check_output(
shlex.split('git rev-parse HEAD'),
cwd=settings.rosetta).strip()
git_diff = subprocess.check_output(
shlex.split('git diff'),
cwd=settings.rosetta).strip()
if benchmark.git_commit != git_commit:
message = "Benchmark \"{0}\" was run with rosetta commit #{1}, but commit #{2} is currently checked out. Press [Ctrl-C] to abort or [Enter] to continue."
message = textwrap.fill(message.format(benchmark.id, benchmark.git_commit[:8], git_commit[:8]))
raw_input(message)
elif benchmark.git_diff != git_diff:
message = "Uncommitted changes have been made to rosetta since benchmark \"{0}\" was run. Press [Ctrl-C] to abort or [Enter] to continue."
message = textwrap.fill(message.format(benchmark.id))
raw_input(message)
# Build the qsub command.
if benchmark.fast:
qsub_command += '-t', '1-{0}'.format((nstruct or 10) * num_pdbs)
qsub_command += '-l', 'h_rt=0:30:00'
else:
qsub_command += '-t', '1-{0}'.format((nstruct or 500) * num_pdbs)
qsub_command += '-l', 'h_rt=4:00:00'
print "Your benchmark \"{0}\" (id={1}) is being resumed".format(
benchmark.name, benchmark_id)
# Submit the job.
utilities.clear_directory('job_output')
qsub_command += '-o', 'job_output', '-e', 'job_output'
subprocess.call(qsub_command + benchmark_command)
from libraries import utilities
from libraries import settings; settings.load(interactive=False)
from libraries import database
# Parse arguments.
if len(sys.argv) != 2 or 'SGE_TASK_ID' not in os.environ:
print 'Usage: SGE_TASK_ID=<id> loop_benchmark.py <benchmark_id>'
sys.exit(1)
task_id = int(os.environ['SGE_TASK_ID']) - 1
benchmark_id = int(sys.argv[1])
# Figure out which loop to benchmark.
with database.connect() as session:
benchmark = session.query(database.Benchmarks).get(benchmark_id)
script_path = benchmark.rosetta_script
script_vars = json.loads(benchmark.rosetta_script_vars or '[]')
flags_path = benchmark.rosetta_flags
fragments_path = benchmark.rosetta_fragments
fast = benchmark.fast
input_pdbs = benchmark.input_pdbs
pdb_path = input_pdbs[task_id % len(input_pdbs)].pdb_path
pdb_tag = os.path.splitext(os.path.basename(pdb_path))[0]
loop_path = re.sub('\.pdb(\.gz)?$', '.loop', pdb_path)
non_random = benchmark.non_random
# Set LD_LIBRARY_PATH so that the MySQL libraries can be found.
rosetta_env = os.environ.copy()
r.name for r in session.execute(
'SELECT DISTINCT name from benchmarks ORDER BY benchmark_id DESC'
)
]
def get_progress(database_name, benchmark_name):
try:
database.test_connect(db_name=database_name)
except RuntimeError, error:
print error
sys.exit(1)
# Create an entry in the benchmarks table.
with database.connect(db_name=database_name) as session:
messages = ['']
# Use the latest benchmark's name if none was supplied
if not benchmark_name:
q = session.query(database.Benchmarks).order_by(
database.Benchmarks.benchmark_id.desc())
if q.count() == 0:
exit(
'There is no benchmark data in the database "{0}".'.format(
database_name))
benchmark_name = q.first().name
messages.append(
'No benchmark was selected. Choosing the most recent benchmark: "{0}".\n'
.format(benchmark_name))
settings.load(interactive=False)
from libraries import database
# Parse arguments.
if len(sys.argv) != 2 or 'SGE_TASK_ID' not in os.environ:
print 'Usage: SGE_TASK_ID=<id> loop_benchmark.py <benchmark_id>'
sys.exit(1)
task_id = int(os.environ['SGE_TASK_ID']) - 1
benchmark_id = int(sys.argv[1])
# Figure out which loop to benchmark.
with database.connect() as session:
benchmark = session.query(database.Benchmarks).get(benchmark_id)
script_path = benchmark.rosetta_script
script_vars = json.loads(benchmark.rosetta_script_vars or '[]')
flags_path = benchmark.rosetta_flags
fragments_path = benchmark.rosetta_fragments
fast = benchmark.fast
input_pdbs = benchmark.input_pdbs
pdb_path = input_pdbs[task_id % len(input_pdbs)].pdb_path
pdb_tag = os.path.splitext(os.path.basename(pdb_path))[0]
loop_path = re.sub('\.pdb(\.gz)?$', '.loop', pdb_path)
non_random = benchmark.non_random
# Set LD_LIBRARY_PATH so that the MySQL libraries can be found.
rosetta_env = os.environ.copy()
def from_database(name_or_id, group_by_name=False):
from libraries import database
from sqlalchemy import desc
with database.connect() as session:
# Decide whether a name or id was used to specify a benchmark run,
# and load the corresponding data out of the database. The meaning
# of name_or_id is inferred from its type: names are expected to be
# strings and ids are expected to be integers. If more than one
# benchmark has the same name, the most recent one will be used.
db_benchmarks = []
try:
id = int(name_or_id)
_db_benchmark = session.query(database.Benchmarks).get(id)
if _db_benchmark is None:
message = "No benchmark '{}' in the database."
utilities.print_error_and_die(message, id)
db_benchmarks = [_db_benchmark]
except ValueError:
name = name_or_id
query = session.query(database.Benchmarks).filter_by(
name=name).order_by(desc(database.Benchmarks.start_time))
db_benchmarks = [q for q in query]
if not group_by_name:
if len(db_benchmarks) > 1:
message = "Multiple benchmarks runs were found with the same name '{0}' (ids are: {1}). If this is expected then set the --group_by_name option.".format(
name,
', '.join(map(str, [b.id for b in db_benchmarks])))
utilities.print_error_and_die(message, name)
b_name = set([db_benchmark.name for db_benchmark in db_benchmarks])
assert (len(b_name) == 1)
b_name = b_name.pop()
b_title = set(
[db_benchmark.title or '' for db_benchmark in db_benchmarks])
if len(b_title) > 1:
colortext.warning(
"There are multiple titles associated with benchmark {0}: '{1}'. Choosing the most recent ('{2}')."
.format(b_name, "', '".join(b_title),
db_benchmarks[0].title or ''))
b_title = db_benchmarks[0].title
else:
b_title = b_title.pop() or None
benchmark = Benchmark(b_name, b_title)
for db_benchmark in db_benchmarks:
# Fill in the benchmark data structure from the database.
print(
"Loading the {0} benchmark (id {1}) from the database...".
format(benchmark.name, db_benchmark.id))
for db_input in db_benchmark.input_pdbs:
path = db_input.pdb_path
if not benchmark.loops.get(path):
benchmark.loops[path] = Loop(benchmark, path)
for structure in db_benchmark.structures:
loop = benchmark.loops[structure.input_tag]
id = len(loop.models) + 1
score = structure.score_features.score
rmsd = structure.rmsd_features.protein_backbone
runtime = structure.runtime_features.elapsed_time
model = Model(loop, id, score, rmsd, runtime)
loop.models.append(model)
return benchmark
def get_benchmark_list_by_name(database_name):
with database.connect(db_name = database_name) as session:
return [r.name for r in session.execute('SELECT DISTINCT name from benchmarks ORDER BY benchmark_id DESC')]
def get_benchmark_list_by_name(database_name):
with database.connect(db_name = database_name) as session:
return [r.name for r in session.execute('SELECT DISTINCT name from benchmarks ORDER BY benchmark_id DESC')]
def get_progress(database_name, benchmark_name):
try: database.test_connect(db_name = database_name)
except RuntimeError, error:
print error
sys.exit(1)
# Create an entry in the benchmarks table.
with database.connect(db_name = database_name) as session:
messages = ['']
# Use the latest benchmark's name if none was supplied
if not benchmark_name:
q = session.query(database.Benchmarks).order_by(database.Benchmarks.benchmark_id.desc())
if q.count() == 0:
exit('There is no benchmark data in the database "{0}".'.format(database_name))
benchmark_name = q.first().name
messages.append('No benchmark was selected. Choosing the most recent benchmark: "{0}".\n'.format(benchmark_name))
# Retrieve the set of benchmark runs associated the benchmark name
q = session.query(database.Benchmarks).filter(database.Benchmarks.name == benchmark_name)
if q.count() == 0:
exit('There is no benchmark data in the database "{0}" for benchmark "{1}".'.format(database_name, benchmark_name))
def from_database(name_or_id, group_by_name = False):
from libraries import database
from sqlalchemy import desc
with database.connect() as session:
# Decide whether a name or id was used to specify a benchmark run,
# and load the corresponding data out of the database. The meaning
# of name_or_id is inferred from its type: names are expected to be
# strings and ids are expected to be integers. If more than one
# benchmark has the same name, the most recent one will be used.
db_benchmarks = []
try:
id = int(name_or_id)
_db_benchmark = session.query(database.Benchmarks).get(id)
if _db_benchmark is None:
message = "No benchmark '{}' in the database."
utilities.print_error_and_die(message, id)
db_benchmarks = [_db_benchmark]
except ValueError:
name = name_or_id
query = session.query(database.Benchmarks).filter_by(name=name).order_by(desc(database.Benchmarks.start_time))
db_benchmarks = [q for q in query]
if not group_by_name:
if len(db_benchmarks) > 1:
message = "Multiple benchmarks runs were found with the same name '{0}' (ids are: {1}). If this is expected then set the --group_by_name option.".format(name, ', '.join(map(str, [b.id for b in db_benchmarks])))
utilities.print_error_and_die(message, name)
b_name = set([db_benchmark.name for db_benchmark in db_benchmarks])
assert(len(b_name) == 1)
b_name = b_name.pop()
b_title = set([db_benchmark.title or '' for db_benchmark in db_benchmarks])
if len(b_title) > 1:
colortext.warning("There are multiple titles associated with benchmark {0}: '{1}'. Choosing the most recent ('{2}').".format(b_name, "', '".join(b_title), db_benchmarks[0].title or ''))
b_title = db_benchmarks[0].title
else:
b_title = b_title.pop() or None
benchmark = Benchmark(b_name, b_title)
for db_benchmark in db_benchmarks:
# Fill in the benchmark data structure from the database.
print "Loading the {0} benchmark (id {1}) from the database...".format(benchmark.name, db_benchmark.id)
for db_input in db_benchmark.input_pdbs:
path = db_input.pdb_path
if not benchmark.loops.get(path):
benchmark.loops[path] = Loop(benchmark, path)
for structure in db_benchmark.structures:
loop = benchmark.loops[structure.input_tag]
id = len(loop.models) + 1
score = structure.score_features.score
rmsd = structure.rmsd_features.protein_backbone
runtime = structure.runtime_features.elapsed_time
model = Model(loop, id, score, rmsd, runtime)
loop.models.append(model)
return benchmark
def complete_benchmark(benchmark_id, nstruct=None):
qsub_command = 'qsub',
benchmark_command = 'loop_benchmark.py', benchmark_id
# You get weird errors if you forget to cast nstruct from string to int.
# Get the progress data for the job
progress_data = get_progress(settings.db_name, benchmark_id)
# Set up nstruct
nstruct = progress_data['nstruct']
if not nstruct:
sys.exit('The nstruct variable is not set for this benchmark. Exiting.')
# Set up the bins for structures that need extra jobs to be run. We run extra jobs in case these fail as well.
bins = {5 : [], 10 : [], 20 : [], 30 : []}
d_bins = bins.keys()
for input_tag, finished_count in progress_data['CountPerStructure'].iteritems():
if finished_count < nstruct:
missing_count = nstruct - finished_count
if missing_count <= 2:
bins[5].append(input_tag)
elif missing_count <= 5:
bins[10].append(input_tag)
elif missing_count <= 10:
bins[20].append(input_tag)
elif missing_count <= 15:
bins[30].append(input_tag)
else:
bin_size = ((int((missing_count - 11)/20.0) + 2) * 20) + 10
bins[bin_size] = bins.get(bin_size, [])
bins[bin_size].append(input_tag)
for d_bin in d_bins:
if not bins[d_bin]:
del bins[d_bin]
with database.connect() as session:
name = benchmark_id
benchmark_records = [r for r in session.query(database.Benchmarks).filter(database.Benchmarks.name == benchmark_id)]
print('')
benchmark_variables = dict(
rosetta_script = set([r.rosetta_script for r in benchmark_records]),
rosetta_script_vars = [json.loads(r.rosetta_script_vars) for r in benchmark_records],
rosetta_flags = set([r.rosetta_flags for r in benchmark_records]),
rosetta_fragments = set([r.rosetta_fragments for r in benchmark_records]),
fast = set([r.fast for r in benchmark_records]),
non_random = set([r.non_random for r in benchmark_records]),
)
for x in range(0, len(benchmark_variables['rosetta_script_vars']) - 1):
if benchmark_variables['rosetta_script_vars'][x] != benchmark_variables['rosetta_script_vars'][x + 1]:
sys.exit('Exception (ambiguity): The benchmark {0} has multiple RosettaScript variable values associated with previous runs: "{1}".'.format(benchmark_id, '", "'.join(map(str, sorted(benchmark_variables['rosetta_script_vars'])))))
for k, v in sorted(benchmark_variables.iteritems()):
if len(v) == 0:
sys.exit('Exception (missing data): The benchmark {0} has no {1} values associated with previous runs.'.format(benchmark_id, k.replace('_', ' ')))
elif k == 'rosetta_script_vars':
benchmark_variables[k] = benchmark_variables[k][0]
elif len(v) > 1:
sys.exit('Exception (ambiguity): The benchmark {0} has multiple {1} values associated with previous runs: "{2}".'.format(benchmark_id, k.replace('_', ' '), '", "'.join(sorted(v))))
else:
benchmark_variables[k] = v.pop()
for nstruct, pdbs in reversed(sorted(bins.iteritems())): # start the longer jobs first
run_benchmark(name, benchmark_variables['rosetta_script'], pdbs, vars=benchmark_variables['rosetta_script_vars'],
flags=benchmark_variables['rosetta_flags'], fragments=benchmark_variables['rosetta_fragments'], nstruct=nstruct,
desc=None, fast=benchmark_variables['fast'], non_random=benchmark_variables['non_random'])