def parameter_selection():
cluster = get_free_nodes()[0]
#cluster = ['dave']
jobs=[]
n=0
is_main_run=0
#filenames=['emotions','yeast','scene','enron','medical','toy10','toy50']#,'cal500','fp','cancer']
filenames=['toy10','toy50','emotions','yeast','scene','enron','medical']
n=0
for filename in filenames:
for graph_type in ['tree']:
for l_norm in ['2']:
in_i=0
for in_c in ['100','75','50','20','10','5','1','0.5','0.25','0.1','0.01']:
in_i+=1
for kth_fold in ['1','2','3','4','5']:
node=cluster[n%len(cluster)]
n+=1
p=multiprocessing.Process(target=singleRSTA, args=(filename,graph_type,'1',node,kth_fold,l_norm,"%d" % in_i,in_c,))
jobs.append(p)
p.start()
time.sleep(2) # fold
pass
time.sleep(2) # c
pass
time.sleep(2*is_main_run) # lnorm
pass
time.sleep(60*is_main_run) # tree
for job in jobs:
job.join()
pass
pass
def run():
logging.info('\t\tGenerating priority queue.')
# get list of files
filelist = os.listdir('../../FeatureExtraction/Results/bids/')
# generate job queueo
jobind = 0
n=1
nodefilelist = []
for filename in filelist:
if checkfile(filename):
continue
if n>5:
job_queue.put((jobind,(nodefilelist)))
jobind+=1
n=1
nodefilelist = []
nodefilelist.append(filename)
n+=1
if len(nodefilelist) > 0:
job_queue.put((jobind,(nodefilelist)))
# get computing node
cluster = get_free_nodes()[0]
#cluster = ['melkinkari'] # if you don't have access to any computer cluster, just use your machine as the only computing node
#cluster = ['ukko133.hpc'] # if you don't have access to any computer cluster, just use your machine as the only computing node
# run jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
is_main_run_factor=1
# running job_queue
threads = []
def run():
cluster = get_free_nodes()[0]
jobs=[]
n=0
filenames=['memeS','memeM','memeL']
filenames=[]
for year in [2000]:
for copaper in [10]:#,10,15]:
for v in [1000]:#,300,500,700,1000,2000]:
filenames.append('%d_%d_%d' % (year,copaper,v))
for filename in filenames:
c='100'
g='0.8'
for nb in ['1','3','5','7']:
for losstype in ['dif','exp']:
if losstype == 'exp':
penaltyrange = ['0.3','0.5','0.7','1','3','5','7']
else:
penaltyrange = ['0.1','0.5','0.8']
for penalty in penaltyrange:
node = cluster[n%len(cluster)]
n+=1
p=multiprocessing.Process(target=singlerun, args=(filename,c,g,nb,penalty,losstype,node,))
jobs.append(p)
p.start()
time.sleep(15)
time.sleep(60)
time.sleep(300)
for job in jobs:
job.join()
def run():
cluster = get_free_nodes()[0]
#cluster = ['dave']
jobs = []
n = 0
is_main_run = 0.01
filenames = [
'emotions', 'yeast', 'scene', 'enron', 'cal500', 'fp', 'cancer',
'medical', 'toy10', 'toy50', 'toy100'
]
n = 0
for filename in filenames:
for graph_type in ['pair', 'tree']:
for t in range(180):
para_t = "%d" % (t + 1)
node = cluster[n % len(cluster)]
n += 1
p = multiprocessing.Process(target=singleMAVAMMlearner,
args=(
filename,
graph_type,
para_t,
node,
))
jobs.append(p)
p.start()
time.sleep(1 * is_main_run)
time.sleep(60 * is_main_run)
for job in jobs:
job.join()
pass
def compute_graph_kernels_in_parallel():
cluster = get_free_nodes()[0]
kernelnamelist = ['lRWkernel','WL', 'WLedge', 'WLspdelta', 'RGkernel1', 'l3graphletkernel', 'untilpRWkernel4', 'untilpRWkernel6', 'untilpRWkernel8', 'untilpRWkernel10', 'spkernel', 'SPkernel', 'RWkernel', 'gestkernel3', 'gestkernel4', 'connectedkernel3', 'connectedkernel4', 'connectedkernel5']
# generate jobs and so on
job_id = 0
job_size = 1
for kernelname in kernelnamelist:
kernelresultfile = "../DTPNCI2015/results/ncicancer_kernel_graph_%s" % (kernelname)
if os.path.exists(kernelresultfile):
continue
job_id = job_id + 1
job_content = kernelname
job_queue.put((job_id, job_content))
logging.info("Processing jobs ...")
# processing jobs
job_size = job_queue.qsize()
logging.info("In total %d jobs" % job_size)
loadpernode = 1
threads = []
counter = 0
for node in cluster:
for i in range(loadpernode):
t = Worker(job_queue, node)
time.sleep(0.5)
counter = counter +1
if counter > job_size:
break
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tThread error!")
for t in threads:
t.join()
pass
def run():
cluster = get_free_nodes()[0]
#cluster = ['dave']
jobs=[]
n=0
is_main_run=1
filenames=['emotions','yeast','scene','enron','cal500','fp','cancer','medical','toy10','toy50','toy100']
filenames=['fp']
n=0
for filename in filenames:
for graph_type in ['pair']:
for t in range(180):
para_t="%d" % (t+1)
node=cluster[n%len(cluster)]
n+=1
p=multiprocessing.Process(target=singleMAMlearner, args=(filename,graph_type,para_t,node,))
jobs.append(p)
p.start()
time.sleep(1*is_main_run)
time.sleep(30*is_main_run)
time.sleep(600*is_main_run)
for job in jobs:
job.join()
pass
def compute_fingerprints_in_parallel():
# cluster is a list of node in an interative cluster e.g. ['ukko001.hpc','ukko002.hpc']
cluster = get_free_nodes()[0]
#cluster = ['melkinkari']
logging.info("Read in list of molecules ...")
moleculelist = []
fin = open('../DTPNCI2015/results/ncicancer_labels')
for line in fin:
moleculelist.append(line.strip())
if fp_flag==1:
fppath = '../structures/FPfiles/'
elif fp_flag==2:
fppath = '../structures/FPfiles/'
elif fp_flag==3:
fppath = '../structures/FPfiles/'
else:
fppath = '../structures/MATLABfiles/'
if not os.path.exists(fppath):
os.makedirs(fppath)
logging.info("Generate jobs ...")
# generate jobs and so on
job_id = 0
job_size = 1
for molecule in moleculelist:
if fp_flag==1:
sdffilename = "%s%s.fp" % (fppath,molecule)
elif fp_flag==2:
sdffilename = "%s%s.fp3" % (fppath,molecule)
elif fp_flag==3:
sdffilename = "%s%s.fp4" % (fppath,molecule)
else:
sdffilename = "%s%s.mat" % (fppath,molecule)
if os.path.exists(sdffilename):
continue
job_id = job_id + 1
job_content=molecule
job_queue.put((job_id,job_content))
logging.info("Processing jobs ...")
# processing jobs
job_size = job_queue.qsize()
logging.info("In total %d jobs" % job_size)
loadpernode = 8
threads = []
counter = 0
for node in cluster:
for i in range(loadpernode):
t = Worker(job_queue, node)
time.sleep(0.5)
counter = counter +1
if counter > job_size:
break
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tThread error!")
for t in threads:
t.join()
pass
def run():
cluster = get_free_nodes()[0]
#cluster = ['dave']
jobs = []
n = 0
is_main_run_factor = 5
filenames = [
'cancer', 'ArD20', 'ArD30', 'toy10', 'toy50', 'emotions', 'yeast',
'medical', 'scene', 'enron', 'cal500', 'fp'
]
n = 0
# generate jobs
for kth_fold in ['1']: #,'2','3','4','5']:
for filename in filenames:
graph_type = 'tree'
for kappa in [180,
280]: #['2','4','8','16','20','32','40','50','60']:
for l_norm in ['2']:
#for t in [5]:#range(0,41,10):
for t in [1, 5] + range(10, 41, 10):
if t == 0:
t = 1
para_t = "%d" % (t)
try:
with open(
"../outputs/%s_%s_%s_f%s_l%s_k%s_pfRSTAs.log"
% (filename, graph_type, para_t, kth_fold,
l_norm, kappa)):
pass
continue
except:
n = n + 1
job_queue.put((n, filename, graph_type, para_t,
kth_fold, l_norm, kappa))
pass # for |T|
pass # for l
pass # for kappa
pass # for datasets
pass # for k fole
# running jobs
job_size = job_queue.qsize()
logging.info("\t\tprocessing %d jobs" % (job_size))
threads = []
for i in range(len(cluster)):
if job_queue.empty():
break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass # def
def compute_fingerprints_in_parallel():
# cluster is a list of node in an interative cluster e.g. ['ukko001.hpc','ukko002.hpc']
cluster = get_free_nodes()[0]
#cluster = ['melkinkari']
logging.info("Read in list of molecules ...")
moleculelist = []
fin = open('../DTPNCI2015/results/ncicancer_labels')
for line in fin:
moleculelist.append(line.strip())
if fp_flag == 1:
fppath = '../structures/FPfiles/'
elif fp_flag == 2:
fppath = '../structures/FPfiles/'
elif fp_flag == 3:
fppath = '../structures/FPfiles/'
else:
fppath = '../structures/MATLABfiles/'
if not os.path.exists(fppath):
os.makedirs(fppath)
logging.info("Generate jobs ...")
# generate jobs and so on
job_id = 0
job_size = 1
for molecule in moleculelist:
if fp_flag == 1:
sdffilename = "%s%s.fp" % (fppath, molecule)
elif fp_flag == 2:
sdffilename = "%s%s.fp3" % (fppath, molecule)
elif fp_flag == 3:
sdffilename = "%s%s.fp4" % (fppath, molecule)
else:
sdffilename = "%s%s.mat" % (fppath, molecule)
if os.path.exists(sdffilename):
continue
job_id = job_id + 1
job_content = molecule
job_queue.put((job_id, job_content))
logging.info("Processing jobs ...")
# processing jobs
job_size = job_queue.qsize()
logging.info("In total %d jobs" % job_size)
loadpernode = 8
threads = []
counter = 0
for node in cluster:
for i in range(loadpernode):
t = Worker(job_queue, node)
time.sleep(0.5)
counter = counter + 1
if counter > job_size:
break
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tThread error!")
for t in threads:
t.join()
pass
def run():
'''
a wrapper function to generate cluster node with good perforamnce, to define a job space, and to launch jobs on computing node of the cluster
'''
logging.info('\t\tGenerating priority queue.')
paramInd = 0
kFold = 5
numLabel = 3200
suffix = 'val'
isTest = '0'
# iterate over the lists
xFilenameList = ['../Data/tcdb.TB', '../Data/tcdb.TICoils', '../Data/tcdb.TIGene3D', '../Data/tcdb.TIHamap', '../Data/tcdb.TIPANTHER', '../Data/tcdb.TIPfam', '../Data/tcdb.TIPhobius', '../Data/tcdb.TIPIRSF', '../Data/tcdb.TIPRINTS', '../Data/tcdb.TIProDom', '../Data/tcdb.TIProSitePatterns', '../Data/tcdb.TIProSiteProfiles', '../Data/tcdb.TISignalP_EUK', '../Data/tcdb.TISignalP_GRAM_NEGATIVE', '../Data/tcdb.TISignalP_GRAM_POSITIVE', '../Data/tcdb.TISMART', '../Data/tcdb.TISUPERFAMILY', '../Data/tcdb.TITIGRFAM', '../Data/tcdb.TITMHMM', '../Data/tcdb.TPSI', '../Data/tcdb.TRPSCDD', '../Data/tcdb.TRPSCDDNCBI', '../Data/tcdb.TRPSCOG', '../Data/tcdb.TRPSKOG', '../Data/tcdb.TRPSPFAM', '../Data/tcdb.TRPSPRK', '../Data/tcdb.TRPSSMART', '../Data/tcdb.TRPSTCDB201509PSSM', '../Data/tcdb.TRPSTIGR']
#xFilenameList = ['../Data/tcdb.TICoils']
yFilenameList = ['../Data/tcdb.TC']
labelIndexList = xrange(1,numLabel+1)
foldIndexList = xrange(1,kFold+1)
# generate job queue, will iterate over c,k,label
for xFilename,yFilename,labelIndex,foldIndex in list(itertools.product(xFilenameList,yFilenameList,labelIndexList,foldIndexList)):
for line in open('parameter_setting'):
words = line.strip().split(' ')
if words[0] == xFilename and words[1] == yFilename:
svmC = words[2]
tmpDir = '../ResultsSVM/%s_%s/' % ( re.sub('.*/','',xFilename), re.sub('.*/','',yFilename))
if not os.path.exists(tmpDir): os.mkdir(tmpDir)
paramInd += 1
outputFilename = tmpDir + '/' + re.sub('.*/','',xFilename) + '_' + re.sub('.*/','',yFilename) + '_l' + str(labelIndex) + '_f' + str(foldIndex) + '_c' +svmC + '_t' + isTest + '_' + suffix
## check if result is ready already
if checkfile(outputFilename): continue
## put parameter into queue
job_queue.put( (paramInd,(str(paramInd),xFilename,yFilename,str(labelIndex),str(foldIndex),svmC,outputFilename,isTest)) )
# get computing node
logging.info('\t\tObtain cluster node')
cluster = get_free_nodes()[0]
#cluster = ['ukko133.hpc']
# run jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
is_main_run_factor=1
# running job_queue
threads = []
workerload = 3
for i in range(len(cluster)):
for j in range(workerload):
if job_queue.empty(): break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass
def run():
is_main_run_factor=5
#filenames=['toy10','toy50','emotions','medical','enron','yeast','scene','cal500','fp','cancer']
#filenames=['cancer']
filenames=['toy10','toy50','emotions','yeast','scene','enron','fp','medical']
n=0
# generate job_queue
logging.info('\t\tGenerating priority queue.')
for newton_method in ['1','0']:
for filename in filenames:
for slack_c in ['1', '10', '0.1']:
for t in [1, 5, 10, 20, 30]:
para_t="%d" % (t)
graph_type = 'tree'
for kappa in ['1','2','3','4','5','6','8','10','12','14','16']:
for l_norm in ['2']:
#for kth_fold in ['1','2','3','4','5']:
for kth_fold in ['1']:
for loss_scaling_factor in ['0.1','1']:
if checkfile(filename,graph_type,para_t,kth_fold,l_norm,kappa,slack_c,loss_scaling_factor,newton_method):
continue
else:
n=n+1
job_queue.put( (n, (filename,graph_type,para_t,kth_fold,l_norm,kappa,slack_c,loss_scaling_factor,newton_method)) )
pass # for newton_method
pass # for loss_scaling_factor
pass # for slack_c
pass # for |T|
pass # for l
pass # for kappa
pass # for datasets
pass # for k fole
# get computing nodes
cluster = get_free_nodes()[0] # if you have access to some interactive computer cluster, get the list of hostnames of the cluster
#cluster = ['melkinkari'] # if you don't have access to any computer cluster, just use your machine as the only computing node
# running job_queue
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
threads = []
for i in range(len(cluster)):
if job_queue.empty():
break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass # def
def run():
logging.info('\t\tGenerating priority queue.')
paramInd = 0
kFold = 5
numLabel = 3200
suffix = 'val'
isTest = '0'
# iterate over the lists
xFilenameList = ['../Data/tcdb.all.KUNIMKL','../Data/tcdb.all.KALIGN','../Data/tcdb.all.KALIGNF', '../Data/tcdb.all.GUNIMKL','../Data/tcdb.all.GALIGN','../Data/tcdb.all.GALIGNF']
yFilenameList = ['../Data/tcdb.TC']
labelIndexList = xrange(1,numLabel+1)
foldIndexList = xrange(1,kFold+1)
# generate job queue, will iterate over c,k,label
for xFilename,yFilename,labelIndex,foldIndex in list(itertools.product(xFilenameList,yFilenameList,labelIndexList,foldIndexList)):
for line in open('parameter_setting'):
words = line.strip().split(' ')
if words[0] == xFilename and words[1] == yFilename:
svmC = words[2]
tmpDir = '../ResultsMKL/%s_%s/' % ( re.sub('.*/','',xFilename), re.sub('.*/','',yFilename))
if not os.path.exists(tmpDir): os.mkdir(tmpDir)
paramInd += 1
outputFilename = tmpDir + '/' + re.sub('.*/','',xFilename) + '_' + re.sub('.*/','',yFilename) + '_l' + str(labelIndex) + '_f' + str(foldIndex) + '_c' +svmC + '_t' + isTest + '_' + suffix
## check if result is ready already
if checkfile(outputFilename): continue
## put parameter into queue
job_queue.put( (paramInd,(str(paramInd),xFilename,yFilename,str(labelIndex),str(foldIndex),svmC,outputFilename,isTest)) )
# get computing node
logging.info('\t\tObtain cluster node')
cluster = get_free_nodes()[0]
#cluster = ['ukko133.hpc']
# run jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
is_main_run_factor=1
# running job_queue
threads = []
workerload = 3
for i in range(len(cluster)):
for j in range(workerload):
if job_queue.empty(): break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass
def run():
jobs=[]
n=0
is_main_run_factor=5
filenames=['cancer','ArD20','ArD30','toy10','toy50','emotions','yeast','medical','scene','enron','cal500','fp']
#filenames=['scene']
n=0
# generate jobs
logging.info('\t\tGenerating job queue.')
for slack_c in ['100','1','0.1','10','0.01','50','0.5','20','0.05','5']:
for kth_fold in ['1','2','3','4','5']:
for filename in filenames:
graph_type = 'tree'
for kappa in ['2','8','16','20']:
for l_norm in ['2']:
for t in range(0,41,10):
if t==0:
t=1
para_t="%d" % (t)
if checkfile(filename,graph_type,para_t,kth_fold,l_norm,kappa,slack_c):
continue
else:
n=n+1
job_queue.put((n,filename,graph_type,para_t,kth_fold,l_norm,kappa,slack_c))
pass # for slack_c
pass # for |T|
pass # for l
pass # for kappa
pass # for datasets
pass # for k fole
# get computing nodes
cluster = get_free_nodes()[0]
# running jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d jobs" % (job_size))
threads = []
for i in range(len(cluster)):
if job_queue.empty():
break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass # def
def parameter_selection():
cluster = get_free_nodes()[0]
#cluster = ['dave']
jobs = []
n = 0
is_main_run = 0
#filenames=['emotions','yeast','scene','enron','medical','toy10','toy50']#,'cal500','fp','cancer']
filenames = [
'toy10', 'toy50', 'emotions', 'yeast', 'scene', 'enron', 'medical'
]
n = 0
for filename in filenames:
for graph_type in ['tree']:
for l_norm in ['2']:
in_i = 0
for in_c in [
'100', '75', '50', '20', '10', '5', '1', '0.5', '0.25',
'0.1', '0.01'
]:
in_i += 1
for kth_fold in ['1', '2', '3', '4', '5']:
node = cluster[n % len(cluster)]
n += 1
p = multiprocessing.Process(target=singleRSTA,
args=(
filename,
graph_type,
'1',
node,
kth_fold,
l_norm,
"%d" % in_i,
in_c,
))
jobs.append(p)
p.start()
time.sleep(2) # fold
pass
time.sleep(2) # c
pass
time.sleep(2 * is_main_run) # lnorm
pass
time.sleep(60 * is_main_run) # tree
for job in jobs:
job.join()
pass
pass
def compute_graph_kernels_in_parallel():
cluster = get_free_nodes()[0]
kernelnamelist = [
'lRWkernel', 'WL', 'WLedge', 'WLspdelta', 'RGkernel1',
'l3graphletkernel', 'untilpRWkernel4', 'untilpRWkernel6',
'untilpRWkernel8', 'untilpRWkernel10', 'spkernel', 'SPkernel',
'RWkernel', 'gestkernel3', 'gestkernel4', 'connectedkernel3',
'connectedkernel4', 'connectedkernel5'
]
# generate jobs and so on
job_id = 0
job_size = 1
for kernelname in kernelnamelist:
kernelresultfile = "../DTPNCI2015/results/ncicancer_kernel_graph_%s" % (
kernelname)
if os.path.exists(kernelresultfile):
continue
job_id = job_id + 1
job_content = kernelname
job_queue.put((job_id, job_content))
logging.info("Processing jobs ...")
# processing jobs
job_size = job_queue.qsize()
logging.info("In total %d jobs" % job_size)
loadpernode = 1
threads = []
counter = 0
for node in cluster:
for i in range(loadpernode):
t = Worker(job_queue, node)
time.sleep(0.5)
counter = counter + 1
if counter > job_size:
break
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tThread error!")
for t in threads:
t.join()
pass
def run():
logging.info('\t\tGenerating priority queue.')
paramind = 0
K = 10
for c in ['0.01','0.05','0.1','0.5','1','5','10','50','100','1000']:
for g in ['0.0001','0.001','0.005','0.01','0.05','0.1','0.5','1','5','10','50','100']:
for k in range(1,(K+1)):
paramind += 1
outfilename = '../../Learning/Results/ParameterSelection/%d' % paramind
if checkfile(outfilename):
continue
job_queue.put((paramind,(str(paramind),str(K),str(k),c,g,outfilename)))
# get computing node
cluster = get_free_nodes()[0]
#cluster = ['ukko133.hpc']
# run jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
is_main_run_factor=1
# running job_queue
threads = []
workerload = 10
for i in range(len(cluster)):
for j in range(workerload):
if job_queue.empty():
break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass # def
def simulationAndRealignment():
'''
#simulation
print "start simulation"
os.system("echo 'start simulation' > %s"%(logFile))
block_size = cpu_num*max_thread
for block_id in range(block):
if (block_id + 1) * block_size > seq_num:
block_size = seq_num - block_id * block_size
os.system("./pair_hmm.py len=%s a=%s e=%s g=%s l=%s rep=%d tree='%s' id=%d in=%s simul=1>> %s"%(options.length, options.indel, options.extension, options.gamma, options.Lambda, block_size, options.tree, block_id, options.ifname, logFile))
'''
#realignment
#get job queue
t_range = 12
print "get the job queue"
job_id = 0
for i in range(t_range):
for j in range(block):
job_queue.put((job_id, 0.05 + 0.05 * i, j))
job_id += 1
#processing
job_size = job_queue.qsize()
print "processing %d jobs" % (job_size)
start = time.time()
os.system("echo 'processing %d jobs' >> %s"%(job_size, logFile))
threads = []
cluster = get_free_nodes()[0]
#cluster=['ukko003.hpc','ukko004.hpc','ukko005.hpc','ukko006.hpc','ukko007.hpc','ukko008.hpc']
load = 1
for i in range(load):
if job_queue.empty():
break
nodes_num = 0
for j in range(max_node):
if job_queue.empty():
break
os.system("echo '---->%d %s'>> %s"%(j, cluster[j%len(cluster)], logFile))
t = Worker(job_queue, cluster[j%len(cluster)])
nodes_num += 1
time.sleep(1)
try:
t.start()
threads.append(t)
except ThreadError:
os.system("echo '\t\tError: thread error caught!'>>%s"%(logFile))
time.sleep(random.randint(5000,6000)/1000.0)
os.system("echo 'using %d nodes in cluster' >>%s"%(nodes_num, logFile))
for t in threads:
t.join()
#combine results
os.system("echo 'combining results' >>%s"%(logFile))
accuracy = []
score = []
for i in range(t_range):
acc = []
lScore = []
for j in range(block):
jobFName = "%s/result_for_%s_%d_with_t%.2f"%(dataDir, os.path.splitext(os.path.basename(options.ifname))[0], j, 0.05+0.05*i)
s = commands.getoutput("tail -1 %s |tr -d '(|)|'|tr ',' ' '"%(jobFName))
s = s.split()
acc.append(float(s[0]))
lScore.append(float(s[1]))
#os.system("rm "+jobFName)
accuracy.append(np.array(acc).mean())
score.append(np.array(lScore).mean())
os.system("echo %s > %s/finalResult"%(' '.join([str(i) for i in accuracy]), dataDir))
os.system("echo %s > %s/finalScore"%(' '.join([str(i) for i in score]), dataDir))
os.system("echo '%.4gs' >> %s"%(time.time()-start, logFile))
def run():
is_main_run_factor = 5
#filenames=['toy10','toy50','emotions','medical','enron','yeast','scene','cal500','fp','cancer']
#filenames=['cancer']
filenames = [
'toy10', 'toy50', 'emotions', 'yeast', 'scene', 'enron', 'fp',
'medical'
]
n = 0
# generate job_queue
logging.info('\t\tGenerating priority queue.')
for newton_method in ['1', '0']:
for filename in filenames:
for slack_c in ['1', '10', '0.1']:
for t in [1, 5, 10, 20, 30]:
para_t = "%d" % (t)
graph_type = 'tree'
for kappa in [
'1', '2', '3', '4', '5', '6', '8', '10', '12',
'14', '16'
]:
for l_norm in ['2']:
#for kth_fold in ['1','2','3','4','5']:
for kth_fold in ['1']:
for loss_scaling_factor in ['0.1', '1']:
if checkfile(filename, graph_type, para_t,
kth_fold, l_norm, kappa,
slack_c, loss_scaling_factor,
newton_method):
continue
else:
n = n + 1
job_queue.put(
(n, (filename, graph_type, para_t,
kth_fold, l_norm, kappa,
slack_c, loss_scaling_factor,
newton_method)))
pass # for newton_method
pass # for loss_scaling_factor
pass # for slack_c
pass # for |T|
pass # for l
pass # for kappa
pass # for datasets
pass # for k fole
# get computing nodes
cluster = get_free_nodes(
)[0] # if you have access to some interactive computer cluster, get the list of hostnames of the cluster
#cluster = ['melkinkari'] # if you don't have access to any computer cluster, just use your machine as the only computing node
# running job_queue
job_size = job_queue.qsize()
logging.info("\t\tProcessing %d job_queue" % (job_size))
threads = []
for i in range(len(cluster)):
if job_queue.empty():
break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass # def
def run():
'''
a wrapper function to generate cluster node with good perforamnce, to define a job space, and to launch jobs on computing node of the cluster
'''
logging.info('\t\tGenerating priority queue.')
paramInd = 0
kFold = 5
suffix = 'sel'
isTest = '0'
# iterate over the lists
xFilenameList = ['../Data/tcdb.all.KUNIMKL','../Data/tcdb.all.KALIGN','../Data/tcdb.all.KALIGNF']
xFilenameList = ['../Data/tcdb.all.GUNIMKL','../Data/tcdb.all.GALIGN','../Data/tcdb.all.GALIGNF']
foldIndexList = xrange(1,kFold+1)
cList = ['1000','5000','10000','50000']
stepSize1List = ['7','9','11']
stepSize2List = ['7','9','11']
cList.reverse()
stepSize1List.reverse()
stepSize2List.reverse()
yFilename = '../Data/tcdb.TC'
EFilename = '../Data/tcdb.TC.E'
SFilename = '../Data/tcdb.TC.SrcSpc'
for xFilename,foldIndex,sopC,stepSize1,stepSize2 in list(itertools.product(xFilenameList,foldIndexList,cList,stepSize1List,stepSize2List)):
tmpDir = '../ResultsSOP/tmp_%s_%s/' % ( re.sub('.*/','',xFilename), re.sub('.*/','',yFilename))
if not os.path.exists(tmpDir): os.mkdir(tmpDir)
paramInd += 1
outputFilename = tmpDir + '/' + re.sub('.*/','',xFilename) + '_' + re.sub('.*/','',yFilename) + '_f' + str(foldIndex) + '_c' +sopC + '_s1' + stepSize1 + '_s2' + stepSize2 + '_t' + isTest + '_' + suffix + '.mat'
logFilename = tmpDir + '/' + re.sub('.*/','',xFilename) + '_' + re.sub('.*/','',yFilename) + '_f' + str(foldIndex) + '_c' +sopC + '_s1' + stepSize1 + '_s2' + stepSize2 + '_t' + isTest + '_' + suffix + '.log'
## check if result is ready already
if checkfile(outputFilename): continue
## put parameter into queue
job_queue.put( (paramInd,(str(paramInd),xFilename,yFilename,EFilename,SFilename,str(foldIndex),sopC,outputFilename,logFilename,stepSize1,stepSize2,isTest,suffix)) )
# get computing node
logging.info('\t\tObtain cluster node')
cluster = get_free_nodes()[0]
#cluster = ['ukko133']
# run jobs
job_size = job_queue.qsize()
logging.info( "\t\tProcessing %d job_queue" % (job_size))
is_main_run_factor=1
# running job_queue
threads = []
workerload = 1
for i in range(len(cluster)):
for j in range(workerload):
if job_queue.empty(): break
t = Worker(job_queue, cluster[i])
time.sleep(is_main_run_factor)
try:
t.start()
threads.append(t)
except ThreadError:
logging.warning("\t\tError: thread error caught!")
pass
for t in threads:
t.join()
pass
pass
