def getSample(self):
cpu_info = OrderedDict()
proc_info = OrderedDict()
uptime_info = {}
nprocs = 0
try:
# Get cpu number
if util.is_exist('/proc/cpuinfo'):
with open('/proc/cpuinfo') as f:
for line in f:
if not line.strip():
nprocs += 1
# Get uptime and idletime
if util.is_exist('/proc/uptime'):
with open('/proc/uptime') as f:
for line in f:
if line.strip():
if len(line.split(' ')) == 2:
uptime_info['uptime'] = {'volume':float(line.split(' ')[0].strip()),'unit':'s'}
uptime_info['idletime'] = {'volume':float(line.split(' ')[1].strip()),'unit':'s'}
uptime_info['cpu_num'] = {'volume':nprocs,'unit':''}
# Compute idle rate
uptime_info['idle_rate'] = {'volume':(uptime_info['idletime']['volume'] * 100) / (uptime_info['cpu_num']['volume'] * uptime_info['uptime']['volume']),'unit':'%'}
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return uptime_info
def build_graph(ids, vocabs, pmi, tfidf):
if util.is_exist("graph.pkl"):
print("loaded")
return util.load_pickle("graph.pkl")
G = networkx.Graph()
G.add_nodes_from(ids)
G.add_nodes_from(vocabs)
cn2 = lambda x: x * (x - 1) // 2
print("Calculating word_word edges")
for (i, w1), (j, w2) in tqdm(combinations(enumerate(vocabs), 2),
total=cn2(len(vocabs))):
if pmi[i][j] > 0:
G.add_edge(w1, w2, weight=pmi[i][j])
print("Calculating doc_word edges")
for i, review_id in tqdm(enumerate(ids), total=len(ids)):
for j, word in enumerate(vocabs):
G.add_edge(review_id, word, weight=tfidf[i][j])
print("Calculating doc_doc edges")
for review_id in tqdm(ids, total=len(ids)):
G.add_edge(review_id, review_id, weight=1)
util.save_pickle(G, "graph.pkl")
return G
def load_data(file, max_num=None):
if util.is_exist("data.pkl"):
print("loaded")
return util.load_pickle("data.pkl")
labels = []
reviews = []
ids = []
with open(file, 'r', encoding='utf-8') as f:
lines = f.readlines()
max_line = len(lines) if max_num is None else max_num + 1
# Ignore Header Line
for line in lines[1:max_line]:
id, label, review = line.split("\t")
labels.append(label)
# remove quotation marks around
review = review[1:len(review) - 1]
reviews.append(review)
# remove quotation marks around
id = id[1:len(id) - 1]
ids.append(id)
reviews = [cleanText(r) for r in reviews]
util.save_pickle((ids, labels, reviews), "data.pkl")
return ids, labels, reviews
def preprocess():
if util.is_exist("preprocessed.pkl"):
print("loading")
return util.load_pickle("preprocessed.pkl")
_, labels, _ = util.load_pickle("data.pkl")
labels = np.array(labels, dtype=np.int32)
G = util.load_pickle("graph.pkl")
print("calc adjacent matrix")
A = networkx.to_numpy_matrix(G, weight="weight")
print("calc degree matrix")
degrees = [d**-0.5 if d != 0 else 0 for _, d in G.degree]
print("normalize adjacent matrix")
'''
degrees = np.diag(degrees)
A_hat = degrees @ A @ degrees
'''
# decrease memory allocation
A_hat = A
for i in tqdm(range(A.shape[0]), total=A.shape[0]):
for j in range(A.shape[1]):
A_hat[i, j] *= degrees[i] * degrees[j]
print("calc feature matrix")
X = np.eye(G.number_of_nodes()) # Features are just identity matrix
util.save_pickle((X, A_hat, labels), "preprocessed.pkl")
return X, A_hat, labels
def __init__(self,
model_type,
input_shape,
look_back,
look_forward,
dataset,
label,
save_path,
weight_path=None):
"""
:param model_type: 생성할 모델 종류
:param input_shape: 입력 데이터 사이즈
:param look_back: 시퀀스의 길이 ( 과거 몇 개의 데이터를 이용해 미래를 예측할 때 '과거 몇 개의 데이터'에 해당하는 길이 )
:param look_forward: 예측하고자하는 값의 범위 ( 과거 몇 개의 데이터를 이용해 미래를 예측할 때 '미래'에 해당하는 길이 )
:param dataset: 전체 데이터셋 ( train, validation, test split하지 않은 전체 데이터셋 )
:param label: 데이터셋 라벨 ( sst, u10, v10 ... )
:param save_path: 학습시 weight, loss curve, prediction 결과를 저장하기 위한 root 폴더
:param weight_path: load할 weight 경로 - 없다면 처음부터 학습
"""
self.model_type = model_type
self.input_shape = input_shape
self.image_shape = input_shape[-3:]
self.look_back = look_back
self.model = self.build_model()
self.model.summary()
self.dataset_arr = dataset_split(dataset, look_back, look_forward)
self.max_list = np.max(self.get_data('train')[0], axis=(0, 1, 2, 3))
self.min_list = np.min(self.get_data('train')[0], axis=(0, 1, 2, 3))
self.save_path = save_path
self.label = label
plot_model(self.model,
os.path.join(self.save_path, 'model.png'),
show_shapes=True)
if weight_path is not None:
is_exist(weight_path, 'weight file')
print("Load weight from {}".format(weight_path))
self.model.load_weights(weight_path)
def calc_tf_idf(reviews, min_df=0.01):
if util.is_exist("tf-idf.pkl"):
print("loaded")
return util.load_pickle("tf-idf.pkl")
vectorizer = TfidfVectorizer(input="content",
stop_words=stopwords.words("english"),
min_df=min_df,
max_df=0.5)
vectorizer.fit(reviews)
tfidf = vectorizer.transform(reviews).toarray()
vocab = vectorizer.get_feature_names()
util.save_pickle((tfidf, vocab), "tf-idf.pkl")
return tfidf, vocab
def getSample(self):
mem_info = OrderedDict()
try:
if util.is_exist('/proc/meminfo'):
#close file is unnecessary
with open('/proc/meminfo') as f:
for line in f:
tmp = line.split(':')
if len(tmp) == 2:
mem_info[tmp[0].strip()] = tmp[1].strip()
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return mem_info
def getSample(self):
mem_info = OrderedDict()
try:
if util.is_exist("/proc/meminfo"):
# close file is unnecessary
with open("/proc/meminfo") as f:
for line in f:
tmp = line.split(":")
if len(tmp) == 2:
mem_info[tmp[0].strip()] = tmp[1].strip()
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return mem_info
def _read_proc_stat(self):
cpu_line = OrderedDict()
f = None
try:
if util.is_exist('/proc/stat'):
f = open('/proc/stat')
lines = f.readlines()
for line in lines:
if line.startswith('cpu'):
tmp = line.strip().split()
cpu_line[tmp[0]] = tmp[1:len(tmp)]
except:
print "Unexpected error: ", sys.exc_info[1]
finally:
if f:
f.close()
return cpu_line
def _read_proc_stat(self):
cpu_line = OrderedDict()
f = None
try:
if util.is_exist('/proc/stat'):
f = open('/proc/stat')
lines = f.readlines()
for line in lines:
if line.startswith('cpu'):
tmp = line.strip().split()
cpu_line[tmp[0]] = tmp[1:len(tmp)]
except:
print "Unexpected error: ", sys.exc_info[1]
finally:
if f:
f.close()
return cpu_line
def getSample(self):
mem_info = OrderedDict()
try:
if util.is_exist('/proc/meminfo'):
#close file is unnecessary
with open('/proc/meminfo') as f:
for line in f:
tmp = line.split(':')
if len(tmp) == 2:
vol_unit = tmp[1].strip().split(' ')
if len(vol_unit) == 2:
tmp_value = self._changeUnit(value=long(vol_unit[0]), force_unit='MB')
elif len(vol_unit) == 1:
tmp_value = {'volume':long(long(vol_unit[0])), 'unit':''}
mem_info[tmp[0].strip()] = tmp_value
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return mem_info
def getSample(self):
net_state = OrderedDict()
title = OrderedDict()
total_item = 0
try:
if util.is_exist('/proc/net/dev'):
with open('/proc/net/dev') as f:
for line in f:
'''
Read items
'''
if line.strip().startswith('Inter'):
tmp = line.strip().split('|')
for i in range(1, len(tmp)):
title[tmp[i].strip()] = []
elif line.strip().startswith('face'):
tmp = line.strip().split('|')
for i in range(1, len(tmp)):
title[title.items()[i-1][0]] = tmp[i].strip().split()
total_item += len(title.items()[i-1][1])
else:
tmp = line.strip().split(':')
tmp_data = OrderedDict()
value = tmp[1].strip().split()
if len(value) == total_item:
cnt = 0
for t_item in title.items():
tmp_data[t_item[0]] = {}
for it in t_item[1]:
tmp_data[t_item[0]][it] = value[cnt]
cnt += 1
else:
print 'number of items error'
net_state[tmp[0]] = tmp_data
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return net_state
def getSample(self):
load_stat = {}
load_info = None
f = None
try:
if util.is_exist('/proc/loadavg'):
f = open('/proc/loadavg')
load_info = f.read().split()
if load_info and len(load_info) == 5:
load_stat['load_1_min'] = {'volume':float(load_info[0]), 'unit':''}
load_stat['load_5_min'] = {'volume':float(load_info[1]), 'unit':''}
load_stat['load_15_min'] = {'volume':float(load_info[2]), 'unit':''}
load_stat['nr_thread'] = load_info[3]
load_stat['last_pid'] = load_info[4]
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
if f:
f.close()
return load_stat
def getSample(self):
load_stat = {}
load_info = None
f = None
try:
if util.is_exist('/proc/loadavg'):
f = open('/proc/loadavg')
load_info = f.read().split()
if load_info and len(load_info) == 5:
load_stat['load_1_min'] = load_info[0]
load_stat['load_5_min'] = load_info[1]
load_stat['load_15_min'] = load_info[2]
load_stat['nr_thread'] = load_info[3]
load_stat['last_pid'] = load_info[4]
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
if f:
f.close()
return load_stat
def calc_pmi(reviews, vocabs, window_size=10):
if util.is_exist("pmi.pkl"):
print("loaded")
return util.load_pickle("pmi.pkl")
word2index = {word: index for index, word in enumerate(vocabs)}
W = 1 # 防止出现p = 1 的情况
W_i = np.zeros(len(vocabs), dtype=np.int32)
W_ij = np.identity(len(vocabs), dtype=np.int32)
vocabs = set(vocabs)
words_list = [[w for w in r.split() if w in vocabs] for r in reviews]
for word_seq in tqdm(words_list, total=len(words_list)):
for i in range(max(len(word_seq) - window_size, 1)):
W += 1
word_set = set(word_seq[i:i + window_size])
for w in word_set:
W_i[word2index[w]] += 1
for w1, w2 in combinations(word_set, 2):
i1 = word2index[w1]
i2 = word2index[w2]
W_ij[i1][i2] += 1
W_ij[i2][i1] += 1
p_i = W_i / W
p_ij = W_ij / W
val = np.zeros(p_ij.shape, dtype=np.float64)
for i in range(len(p_i)):
for j in range(len(p_i)):
if p_ij[i, j] != 0 and p_i[i] * p_i[j] != 0:
val[i, j] = math.log(p_ij[i, j] / (p_i[i] * p_i[j]))
util.save_pickle(val, "pmi.pkl")
return val
def getSample(self):
cpu_info = OrderedDict()
proc_info = OrderedDict()
nprocs = 0
try:
if util.is_exist('/proc/cpuinfo'):
with open('/proc/cpuinfo') as f:
for line in f:
if not line.strip():
cpu_info['proc%s' % nprocs] = proc_info
nprocs += 1
proc_info = OrderedDict()
else:
if len(line.split(':')) == 2:
proc_info[line.split(':')[0].strip()] = line.split(':')[1].strip()
else:
proc_info[line.split(':')[0].strip()] = ''
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return cpu_info
def getSample(self):
cpu_info = OrderedDict()
proc_info = OrderedDict()
nprocs = 0
try:
if util.is_exist('/proc/cpuinfo'):
with open('/proc/cpuinfo') as f:
for line in f:
if not line.strip():
cpu_info['proc%s' % nprocs] = proc_info
nprocs += 1
proc_info = OrderedDict()
else:
if len(line.split(':')) == 2:
proc_info[line.split(':')[0].strip(
)] = line.split(':')[1].strip()
else:
proc_info[line.split(':')[0].strip()] = ''
except:
print "Unexpected error:", sys.exc_info()[1]
finally:
return cpu_info
from ocean_predictor import OceanPredictor
from keras.optimizers import Adam
from util import load_data, is_exist
import numpy as np
import json
import os
import time
import shutil
if __name__ == '__main__':
hyperparameter_json_path = './hyperparameters.json'
is_exist(hyperparameter_json_path, 'hyperparameter json file')
with open(hyperparameter_json_path) as json_file:
"""
주의 - json은 문자를 표현할 때 작은 따옴표(')를 못 씀, 큰 따옴표(")만 사용 가능함
Hyperparameter는 hyperparameter_json_path에 위치한 json파일 내부 변수에서 세팅하면 됨
null로 주어져도 되는 것들 (python에서 로드될 때는 null이 None으로 변환됨)
exp_name => null로 주어질 경우 현재 시간을 root 폴더 이름으로 사용함
weight_path => null로 주어질 경우 학습을 처음부터 수행함
random_seed => null로 주어질 경우 weight initialize가 매 학습마다 다르게 됨
학습 할 때마다 exp_name 내부에 다음과 같은 파일들이 만들어짐
exp_name ---- | SST ---- | train, test 데이터셋에 대한 예측값, 실제값 비교 그래프
---- | SWH ---- | train, test 데이터셋에 대한 예측값, 실제값 비교 그래프
---- | ... ( 다른 label 데이터들 )
---- | ... ( 다른 label 데이터들 )
---- | ... ( 다른 label 데이터들 )
