def train():
'''
$1 path to config file
'''
start = datetime.now()
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',
level=logging.INFO)
# loading configs;
config = get_config(sys.argv[2])
k = config['k']
nthread = config['nthread']
asyn = config['asyn']
mm_path = config['mm_path']
var_path = config['var_path']
minibatch = config['minibatch']
corpus = _mCorpus.get_corpus(mm_path)
V = corpus.num_terms
if asyn:
eta = _masynf.asyn_framework(corpus, k, V, nthread, minibatch,
var_path)
else:
eta = _msynf.syn_framework(corpus, k, V, nthread, minibatch, var_path,
True)
fn = 'eta.final.pickle'
path = os.path.join(var_path, fn)
_mea.write_eta(eta, path)
end = datetime.now()
print end - start
def train():
'''
$1 path to config file
'''
start = datetime.now()
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.INFO)
# loading configs;
config = get_config(sys.argv[2])
k = config['k']
nthread = config['nthread']
asyn = config['asyn']
mm_path=config['mm_path']
var_path = config['var_path']
minibatch = config['minibatch']
corpus = _mCorpus.get_corpus(mm_path)
V = corpus.num_terms
if asyn:
eta = _masynf.asyn_framework(corpus,k,V,nthread,minibatch,var_path)
else:
eta = _msynf.syn_framework(corpus,k,V,nthread,minibatch,var_path,True)
fn = 'eta.final.pickle'
path = os.path.join(var_path,fn)
_mea.write_eta(eta,path)
end = datetime.now()
print end-start
def main():
# Initializations and preliminaries
comm = MPI.COMM_WORLD # get MPI communicator object
size = comm.size # total number of processes
rank = comm.rank # rank of this process
status = MPI.Status() # get MPI status object
tags = enum('READY', 'DONE', 'EXIT', 'START')
if rank == 0:
# Master process
'''
$1 path to config file
'''
start = datetime.now()
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s',
level=logging.INFO)
# loading configs;
config = get_config(sys.argv[2])
k = config['k']
nthread = config['nthread']
asyn = config['asyn'] # here, the value should be 'mpi'
mm_path = config['mm_path']
var_path = config['var_path']
minibatch = config['minibatch']
corpus = _mCorpus.get_corpus(mm_path)
V = corpus.num_terms
eta = master_process(comm, status, tags, corpus, k, V, nthread,
minibatch, var_path)
# store the final pickle
fn = 'eta.final.pickle'
path = os.path.join(var_path, fn)
_mea.write_eta(eta, path)
end = datetime.now()
print end - start
else:
# Worker process
name = MPI.Get_processor_name()
worker_process(comm, status, tags, name)
def main():
# Initializations and preliminaries
comm = MPI.COMM_WORLD # get MPI communicator object
size = comm.size # total number of processes
rank = comm.rank # rank of this process
status = MPI.Status() # get MPI status object
tags = enum('READY', 'DONE', 'EXIT', 'START')
if rank == 0:
# Master process
'''
$1 path to config file
'''
start = datetime.now()
logging.basicConfig(format='%(asctime)s : %(levelname)s : %(message)s', level=logging.INFO)
# loading configs;
config = get_config(sys.argv[2])
k = config['k']
nthread = config['nthread']
asyn = config['asyn'] # here, the value should be 'mpi'
mm_path=config['mm_path']
var_path = config['var_path']
minibatch = config['minibatch']
corpus = _mCorpus.get_corpus(mm_path)
V = corpus.num_terms
eta = master_process(comm,status,tags,corpus,k,V,nthread,minibatch,var_path)
# store the final pickle
fn = 'eta.final.pickle'
path = os.path.join(var_path,fn)
_mea.write_eta(eta,path)
end = datetime.now()
print end-start
else:
# Worker process
name = MPI.Get_processor_name()
worker_process(comm,status,tags,name)
def write_eta(self, path):
self.lock.acquire()
_mea.write_eta(self.eta, path)
self.lock.release()
def write_eta(self,path):
self.lock.acquire()
_mea.write_eta(self.eta,path)
self.lock.release()
def syn_framework(corpus,
k,
V,
nthread,
minibatch,
var_path,
record_eta=False):
# configs
thread_batch = minibatch / nthread
# ids
doc_id = 0
batch_id = 0
round_id = 0
# temp data
doc_buffer = []
batch_buffer = [] # [(docs,etas)]
voc_temp = set()
# global data
eta = {}
for doc in corpus:
for vid, count in doc:
voc_temp.add(vid)
doc_buffer.append(doc)
if doc_id % thread_batch == thread_batch - 1:
eta_temp = _mea.get_eta(k, eta, voc_temp)
etaSum = _mea.get_eta_sum(eta, k, V)
batch_buffer.append((doc_buffer, eta_temp, etaSum))
# clear doc buffer
doc_buffer = []
voc_temp = set()
if batch_id % nthread == nthread - 1:
# update eta
eta = syn_master(batch_buffer, k, nthread, eta,
_mea.get_alpha(k))
if record_eta:
fn = 'eta.{}.pickle'.format(round_id)
path = os.path.join(var_path, fn)
_mea.write_eta(eta, path)
# clear batch_buffer
batch_buffer = []
round_id += 1
logging.info('round:{}, batch:{}'.format(round_id, batch_id))
batch_id += 1
doc_id += 1
# process the docs in current doc_buffer
if len(doc_buffer) > 0:
# form a new batch
eta_temp = _mea.get_eta(k, eta, voc_temp)
etaSum = _mea.get_eta_sum(eta, k, V)
batch_buffer.append((doc_buffer, eta_temp, etaSum))
# form a new round
eta = syn_master(batch_buffer, k, len(batch_buffer), eta,
_mea.get_alpha(k))
if record_eta:
fn = 'eta.{}.pickle'.format(round_id)
path = os.path.join(var_path, fn)
_mea.write_eta(eta, path)
round_id += 1
batch_id += 1
return eta
def syn_framework(corpus,k,V,nthread,minibatch,var_path,record_eta = False):
# configs
thread_batch = minibatch/nthread
# ids
doc_id = 0
batch_id = 0
round_id = 0
# temp data
doc_buffer = []
batch_buffer = [] # [(docs,etas)]
voc_temp = set()
# global data
eta = {}
for doc in corpus:
for vid,count in doc:
voc_temp.add(vid)
doc_buffer.append(doc)
if doc_id % thread_batch == thread_batch - 1:
eta_temp = _mea.get_eta(k,eta, voc_temp)
etaSum = _mea.get_eta_sum(eta,k,V)
batch_buffer.append((doc_buffer,eta_temp,etaSum))
# clear doc buffer
doc_buffer = []
voc_temp = set()
if batch_id % nthread == nthread - 1:
# update eta
eta = syn_master(batch_buffer,k,nthread,eta,_mea.get_alpha(k))
if record_eta:
fn = 'eta.{}.pickle'.format(round_id)
path = os.path.join(var_path,fn)
_mea.write_eta(eta,path)
# clear batch_buffer
batch_buffer = []
round_id += 1
logging.info('round:{}, batch:{}'.format(round_id,batch_id))
batch_id += 1
doc_id += 1
# process the docs in current doc_buffer
if len(doc_buffer) > 0:
# form a new batch
eta_temp = _mea.get_eta(k,eta, voc_temp)
etaSum = _mea.get_eta_sum(eta,k,V)
batch_buffer.append((doc_buffer,eta_temp,etaSum))
# form a new round
eta = syn_master(batch_buffer,k,len(batch_buffer),eta,_mea.get_alpha(k))
if record_eta:
fn = 'eta.{}.pickle'.format(round_id)
path = os.path.join(var_path,fn)
_mea.write_eta(eta,path)
round_id +=1
batch_id +=1
return eta