def run(self):
hdfs = luigi.contrib.hdfs.hadoopcli_clients.create_hadoopcli_client()
dates = self.version_date()
if len(dates) == 0:
raise Exception("These's no user data in[%s]" % self.input()[0].fn)
exit_code = -1
mr_path = os.path.dirname(self.bin)
copyfile("%s/mapreduce.properties.user" % mr_path, "%s/mapreduce.properties" % mr_path)
if not os.path.exists(self.local_version):
hdfs.remove(self.user_root)
hdfs.mkdir(self.user_root)
exit_code = mr_cmd(self.bin, "pr.user")
if exit_code != 0 or not check_mr_success(self.merge):
raise Exception("GetExternalUser failed")
hdfs.rename(self.merge, self.output()["user"].path)
else:
local_dates = [line.strip() for line in open(self.local_version)]
latest_dates = set(dates) - set(local_dates)
if len(latest_dates) == 0:
raise Exception("These's no new arrival user data")
hdfs.remove(self.user_root)
hdfs.mkdir(self.user_root)
hdfs.mkdir(self.increamental_archive)
for d in latest_dates:
hdfs.copy("%s/%s" % (self.input()[0].path, d), "%s/%s" % (self.increamental_archive, d))
exit_code = mr_cmd(self.bin, "pr.latest.user")
if exit_code != 0 or not check_mr_success(self.output()["user"].path):
raise Exception("GetExternalUser failed")
# make version tag
with self.output()["version"].open("w") as version_fd:
dates.sort()
for d in dates:
print >> version_fd, d
def run(self):
hdfs = luigi.contrib.hdfs.hadoopcli_clients.create_hadoopcli_client()
hdfs.remove(self.paper_root)
hdfs.mkdir(self.paper_root)
mr_path = os.path.dirname(self.bin)
copyfile("%s/mapreduce.properties.doc" % mr_path, "%s/mapreduce.properties" % mr_path)
exit_code = mr_cmd(self.bin, "pr.paper")
if exit_code != 0 or not check_mr_success(self.output().path):
raise Exception("GetExternalPaper failed")
def infer_topic(in_fn, model_fn, out_fn, conf):
parser = SafeConfigParser()
parser.read(conf)
root = parser.get("basic", "root")
hadoop_stream = parser.get('basic', 'hadoop_stream')
topic_num = parser.getint('plda+', 'topic_num')
alpha = 50.0 / topic_num
task_id = uuid.uuid4()
infer_in_path = "%s/%s" % (parser.get('plda+', 'infer_in_path'), task_id)
infer_out_path = "%s/%s" % (parser.get('plda+', 'infer_out_path'), task_id)
infer_burn_in_iter = parser.getint('plda+', 'infer_burn_in_iter')
infer_total_iter = parser.getint('plda+', 'infer_total_iter')
infer_reduce_tasks = parser.getint('plda+', 'infer_reduce_tasks')
infer_reducer_mb = parser.getint('plda+', 'infer_reducer_mb')
mapper = '%s/plda/infer_mapper' % root
reducer = '%s/plda/infer_reducer' % root
reducer_wrapper = '%s/data/temp/reducer_wrapper.sh' % root
hdfs = luigi.contrib.hdfs.hadoopcli_clients.create_hadoopcli_client()
hdfs.mkdir(infer_in_path)
hdfs.put(in_fn, infer_in_path)
with open(reducer_wrapper, 'w') as wrapper_fd:
print >> wrapper_fd, "#!/bin/bash"
print >> wrapper_fd, "./infer_reducer --alpha %f --beta 0.01 --model_file ./%s --burn_in_iterations %d --total_iterations %d -sparse true" % \
(alpha, os.path.basename(model_fn), infer_burn_in_iter, infer_total_iter)
cmd = '''hadoop jar %s \
-D mapred.job.name="mr plda+ infer" \
-D mapred.job.map.memory.mb=32 \
-D mapred.job.reduce.memory.mb=%d \
-D io.compression.codecs=org.apache.hadoop.io.compress.DefaultCodec \
-input %s \
-output %s \
-file %s \
-file %s \
-file %s \
-file %s \
-mapper ./infer_mapper \
-reducer ./reducer_wrapper.sh \
-numReduceTasks %d
'''
cmd = cmd % (hadoop_stream, infer_reducer_mb,
infer_in_path, infer_out_path,
model_fn, mapper, reducer, reducer_wrapper,
infer_reduce_tasks)
os.system(cmd)
os.remove(reducer_wrapper)
if check_mr_success(infer_out_path):
with open(out_fn, 'w') as out_fd:
get_mr_dir(infer_out_path, out_fd)
hdfs.remove(infer_in_path)
hdfs.remove(infer_out_path)
else:
hdfs.remove(infer_in_path)
hdfs.remove(infer_out_path)
raise Exception("failed to infer topic")