def run_iter(json_file, init_model):
prev_model = init_model
fp = open(json_file, 'r')
jdata = json.load(fp)
numb_iter = jdata["numb_iter"]
numb_task = 8
record = "record.rit"
iter_rec = [0, -1]
if os.path.isfile(record):
with open(record) as frec:
for line in frec:
iter_rec = [int(x) for x in line.split()]
logging.info("continue from iter %03d task %02d" %
(iter_rec[0], iter_rec[1]))
global exec_machine
for ii in range(numb_iter):
if ii > 0:
prev_model = glob.glob(
make_iter_name(ii - 1) + "/" + train_name + "/*pb")
for jj in range(numb_task):
if ii * max_tasks + jj <= iter_rec[0] * max_tasks + iter_rec[1]:
continue
if jj == 0:
log_iter("make_temp", ii, jj)
make_temp(ii, json_file, prev_model)
elif jj == 1:
log_iter("run_temp", ii, jj)
run_temp(ii, json_file)
elif jj == 2:
log_iter("post_temp", ii, jj)
cont = post_temp(ii, json_file)
if not cont:
log_iter("no more conf needed", ii, jj)
return
elif jj == 3:
log_iter("make_res", ii, jj)
cont = make_res(ii, json_file)
elif jj == 4:
log_iter("run_res", ii, jj)
run_res(ii, json_file, exec_machine)
elif jj == 5:
log_iter("post_res", ii, jj)
post_res(ii, json_file)
elif jj == 6:
log_iter("make_train", ii, jj)
make_train(ii, json_file)
elif jj == 7:
log_iter("run_train", ii, jj)
run_train(ii, json_file, exec_machine)
else:
raise RuntimeError("unknow task %d, something wrong" % jj)
record_iter(record, ii, jj)
def run_iter(json_file, init_model):
base_dir = os.getcwd()
prev_model = init_model
fp = open(json_file, 'r')
jdata = json.load(fp)
sits_param = jdata.get("sits_settings", None)
numb_iter = jdata["numb_iter"]
niter_per_sits = sits_param.get("niter_per_sits", 100000000)
numb_task = 8
record = "record.train"
record_sits = "record.sits"
cleanup = jdata["cleanup"]
iter_rec = [0, -1]
sits_iter_rec = [0, -1]
if os.path.isfile(record):
with open(record) as frec:
for line in frec:
iter_rec = [int(x) for x in line.split()]
logging.info("continue from iter %03d task %02d" %
(iter_rec[0], iter_rec[1]))
if os.path.isfile(record_sits):
with open(record_sits) as frec:
for line in frec:
sits_iter_rec = [int(x) for x in line.split()]
logging.info("continue from iter %03d task %02d" %
(sits_iter_rec[0], sits_iter_rec[1]))
global exec_machine
bPost_train = jdata.get("post_train")
if sits_iter_rec == [0, -1]:
create_path("sits")
data_name = "data"
for ii in range(iter_rec[0], numb_iter):
if ii > 0:
prev_model = glob.glob(
make_iter_name(ii - 1) + "/" + train_name + "/*pb")
train_ori(iter_index=ii, json_file=json_file)
record_iter(record, ii, 0)
def run_iter(json_file, init_model):
base_dir = os.getcwd()
prev_model = init_model
fp = open(json_file, 'r')
jdata = json.load(fp)
sits_param = jdata.get("sits_settings", None)
numb_iter = jdata["numb_iter"]
niter_per_sits = sits_param.get("niter_per_sits", 100000000)
numb_task = 8
record = "record.rid"
record_sits = "record.sits"
cleanup = jdata["cleanup"]
iter_rec = [0, -1]
sits_iter_rec = [0, -1]
if os.path.isfile(record):
with open(record) as frec:
for line in frec:
iter_rec = [int(x) for x in line.split()]
logging.info("continue from iter %03d task %02d" %
(iter_rec[0], iter_rec[1]))
if os.path.isfile(record_sits):
with open(record_sits) as frec:
for line in frec:
sits_iter_rec = [int(x) for x in line.split()]
logging.info("continue from iter %03d task %02d" %
(sits_iter_rec[0], sits_iter_rec[1]))
global exec_machine
bPost_train = jdata.get("post_train")
if sits_iter_rec == [0, -1]:
create_path("sits")
data_name = "data"
for ii in range(iter_rec[0], numb_iter):
kk = int(ii / niter_per_sits)
data_name = "data%03d" % (kk + 1)
if ii > 0:
prev_model = glob.glob(
make_iter_name(ii - 1) + "/" + train_name + "/*pb")
if ii % niter_per_sits == 0:
log_iter("run_sits_iter", kk, 0)
if not os.path.exists(join("sits", make_iter_name(kk))):
create_path(join("sits", make_iter_name(kk)))
if kk > 0:
open(join("sits", make_iter_name(kk - 1), "rid_iter_end.dat"),
"w+").write("%d" % ii)
open(join("sits", make_iter_name(kk), "rid_iter_begin.dat"),
"w+").write("%d" % ii)
for jj in range((sits_iter_rec[1] + 1) % 6, 6):
if kk * max_tasks + jj <= sits_iter_rec[
0] * max_tasks + sits_iter_rec[1]:
continue
os.chdir(base_dir)
if jj == 0:
make_sits_iter(kk, json_file, prev_model)
elif jj == 1:
run_sits_iter(kk, json_file)
elif jj == 2:
post_sits_iter(kk, json_file)
elif jj == 3:
if kk > 0:
make_train_eff(kk, json_file)
elif jj == 4:
if kk > 0:
run_train_eff(kk, json_file, exec_machine)
elif jj == 5:
if kk > 0:
post_train_eff(kk, json_file)
record_iter(record_sits, kk, jj)
for jj in range(numb_task):
if ii * max_tasks + jj <= iter_rec[0] * max_tasks + iter_rec[1]:
continue
os.chdir(base_dir)
if jj == 0:
log_iter("make_enhc", ii, jj)
# logging.info ("use prev model " + str(prev_model))
make_enhc(ii, json_file, prev_model)
elif jj == 1:
log_iter("run_enhc", ii, jj)
run_enhc(ii, json_file)
elif jj == 2:
log_iter("post_enhc", ii, jj)
post_enhc(ii, json_file)
elif jj == 3:
log_iter("make_res", ii, jj)
cont = make_res(ii, json_file)
if not cont:
log_iter("no more conf needed", ii, jj)
return
elif jj == 4:
log_iter("run_res", ii, jj)
run_res(ii, json_file, exec_machine)
elif jj == 5:
log_iter("post_res", ii, jj)
post_res(ii, json_file, data_name=data_name)
elif jj == 6:
log_iter("make_train", ii, jj)
make_train(ii, json_file, data_name=data_name)
elif jj == 7:
log_iter("run_train", ii, jj)
run_train(ii, json_file, exec_machine, data_name=data_name)
if cleanup:
clean_train(ii)
clean_enhc(ii)
clean_enhc_confs(ii)
clean_res(ii)
else:
raise RuntimeError("unknow task %d, something wrong" % jj)
record_iter(record, ii, jj)