def parse_dir(dir_name):
res_class, res_sem = [], []
f_s = open(dir_name+"/s_test.txt", "r")
f_s_lines = f_s.readlines()
f_s.close()
style_type = [int(x) for x in f_s_lines]
print("dir_name model_type \t transfer_strength content_reservation mixture")
print("================================================================================")
#model_types = ["multi_decoder", "embedding", "memory"]
model_types = get_sub_dirnames(dir_name)
model_types = [i for i in model_types if not i.endswith("txt")]
for model_type in model_types:
model_name = dir_name+"/"+model_type+'/'
score_class = parse_model_class(model_name, style_type)
score_sem = parse_model_sem(model_name, style_type)
#score_sem = (score_sem-0.828)/0.172
if model_type=="memory":
print(dir_name, "\t", model_type, "\t\t\t", score_class, "\t\t", score_sem, "\t", score_class*score_sem/(score_class+score_sem))
else:
print(dir_name, "\t", model_type, "\t\t", score_class, "\t\t", score_sem, "\t", score_class*score_sem/(score_class+score_sem))
res_class.append(score_class)
res_sem.append(score_sem)
print("===============================================================================")
return res_class, res_sem
def gen_score(test_dir_name):
emb = Embedding(100)
word_dict = emb.get_all_emb()
subdir_names = ['multi_decoder', 'embedding', 'memory']
subdir_names = get_sub_dirnames(test_dir_name)
subdir_names = [i for i in subdir_names if not i.endswith("txt")]
for dir_name in subdir_names:
for index_name in ["0", "1"]:
q_file = test_dir_name+"/q_test.txt"
r_file = test_dir_name+"/"+dir_name+"/style"+index_name+".txt"
w_file = test_dir_name+"/"+dir_name+"/style"+index_name+"_semantics.txt"
com_file(q_file, r_file, w_file, word_dict)
return
def get_sample_data(dir_name):
workbook = xlsxwriter.Workbook('toscore.xlsx')
worksheet = workbook.add_worksheet()
subdir_names = get_sub_dirnames(dir_name)
subdir_names = [i for i in subdir_names if not i.endswith("txt")]
q, r = [], []
subdir_names = ["embedding"]
for sub_dir_name in subdir_names:
f0 = open(dir_name + "/" + sub_dir_name + "/style0.txt").readlines()
f1 = open(dir_name + "/" + sub_dir_name + "/style1.txt").readlines()
f_q = open(dir_name + "/q_test.txt").readlines()
f_s = open(dir_name + "/s_test.txt").readlines()
for i, j in enumerate(f0):
if int(f_s[i].strip()) == 0:
continue
else:
q.append(f_q[i])
r.append(j)
random.seed(1027)
random.shuffle(q)
random.seed(1027)
random.shuffle(r)
i = -1
count = 0
cc = 0
while True:
i += 1
if q[i].strip() == r[i].strip():
cc += 1
#continue
worksheet.write(count * 4, 0, "news")
worksheet.write(count * 4 + 1, 0, "paper")
worksheet.write(count * 4 + 2, 0, "score")
worksheet.write(count * 4, 1, q[i].strip().decode('utf-8'))
worksheet.write(count * 4 + 1, 1, r[i].strip().decode('utf-8'))
worksheet.write(count * 4 + 2, 1, "")
count += 1
if count > 99:
break
print(cc)
workbook.close()
def get_sample_data(dir_name):
workbook = xlsxwriter.Workbook('toscore1.xlsx')
worksheet = workbook.add_worksheet()
subdir_names = get_sub_dirnames(dir_name)
subdir_names = [i for i in subdir_names if not i.endswith("txt")]
q, r = [], []
for sub_dir_name in subdir_names:
f0 = open(dir_name + "/" + sub_dir_name + "/style0.txt").readlines()
f1 = open(dir_name + "/" + sub_dir_name + "/style1.txt").readlines()
f_q = open(dir_name + "/q_test.txt").readlines()
f_s = open(dir_name + "/s_test.txt").readlines()
for i, j in enumerate(f0):
if int(f_s[i].strip()) == 1:
continue
else:
q.append(f_q[i])
r.append(j)
random.seed(1027)
random.shuffle(q)
random.seed(1027)
random.shuffle(r)
i = -1
count = 0
cc = 0
while True:
i += 1
if q[i].strip() == r[i].strip():
cc += 1
#continue
res0 = q[i].strip()
res1 = r[i].strip()
worksheet.write(count, 0, res0.decode('utf-8'))
worksheet.write(count, 1, res1.decode('utf-8'))
count += 1
if count > 99:
break
print cc
workbook.close()
def test(test_dir_name):
model = model_from_json(open('./model/model_architecture.json').read())
model.load_weights('./model/model_weights.h5')
#subdir_names = ['multi_decoder', 'embedding', 'memory']
subdir_names = get_sub_dirnames(test_dir_name)
subdir_names = [i for i in subdir_names if not i.endswith("txt")]
for dir_name in subdir_names:
for index_name in ["0", "1"]:
test_x = get_test_data(test_dir_name + "/"+ dir_name +"/style"+ index_name +".txt")
test_x = sequence.pad_sequences(test_x, maxlen=maxlen)
print(("test_x shape:"+str(test_x.shape)))
scores=model.predict_proba(test_x)
f = open(test_dir_name + "/"+ dir_name +"/style"+ index_name +"_classification.txt", "w")
for score in scores:
f.write(str(score[0])+"\n")
f.close()
return