def upload_file():
"""
Lấy dữ liệu ảnh từ phía browser về server
Thực hiện nhận dạng gương mặt
:return: Trang web nhận dạng gương mặt
"""
if request.method == 'POST':
utils.reload()
# Kiểm tra xem có đối tượng files trong trang 'recognition.html' không
if 'file' not in request.files:
flash('No file part')
return redirect(request.url)
image = request.files['file']
# Trường hợp người dùng chưa tải ảnh lên
# Website không thay đổi
if image.filename == '':
flash('No selected file')
return redirect(request.url)
# Trường hợp người dùng đã tải ảnh lên
# Lấy dữ liệu ảnh về server và lưu trũ để xử lý
if image and allowed_file(image.filename):
filename = secure_filename(image.filename)
image.save(os.path.join(app.config['UPLOAD_FOLDER'], filename))
# Thực hiện nhận dạng gương mặt có trong ảnh
utils.face_recog('static/' + filename)
#utils.recog_LBPH('static/' + filename)
return render_template("recognition.html", image_upload=filename)
return render_template("recognition.html")
def configure_haproxy(service_ports):
'''
Configure HAProxy based on the current peers in the service
cluster using the provided port map:
"swift": [ 8080, 8070 ]
HAproxy will also be reloaded/started if required
service_ports: dict: dict of lists of [ frontend, backend ]
'''
cluster_hosts = {}
cluster_hosts[os.getenv('JUJU_UNIT_NAME').replace('/', '-')] = \
unit_get('private-address')
for r_id in relation_ids('cluster'):
for unit in relation_list(r_id):
cluster_hosts[unit.replace('/', '-')] = \
relation_get(attribute='private-address',
rid=r_id,
unit=unit)
context = {'units': cluster_hosts, 'service_ports': service_ports}
with open(HAPROXY_CONF, 'w') as f:
f.write(render_template(os.path.basename(HAPROXY_CONF), context))
with open(HAPROXY_DEFAULT, 'w') as f:
f.write('ENABLED=1')
reload('haproxy')
def _build_data(self, data_dir, data_dir_new, cfg, db):
data_filename = data_dir + '/' + cfg.data_file
with open(data_filename, 'r') as f:
origin_data = json.load(f)
for part in ['train','valid','test']:
self.data[part] = []
self.goal[part] = {}
valList = []
with open(data_dir + '/' + cfg.val_file) as f:
for line in f:
valList.append(line.split('.')[0])
testList = []
with open(data_dir + '/' + cfg.test_file) as f:
for line in f:
testList.append(line.split('.')[0])
for k_sess in origin_data:
sess = origin_data[k_sess]
if k_sess in valList:
part = 'valid'
elif k_sess in testList:
part = 'test'
else:
part = 'train'
turn_data, session_data = init_session(k_sess, cfg)
belief_state = turn_data['belief_state']
goal_state = turn_data['goal_state']
init_goal(session_data, goal_state, sess['goal'], cfg)
self.goal[part][k_sess] = deepcopy(session_data)
current_domain = ''
book_domain = ''
turn_data['trg_user_action'] = {}
turn_data['trg_sys_action'] = {}
for i, turn in enumerate(sess['log']):
turn_data['others']['turn'] = i
turn_data['others']['terminal'] = i + 2 >= len(sess['log'])
da_origin = turn['dialog_act']
expand_da(da_origin)
turn_data['belief_state'] = deepcopy(belief_state) # from previous turn
turn_data['goal_state'] = deepcopy(goal_state)
if i % 2 == 0: # user
turn_data['sys_action'] = deepcopy(turn_data['trg_sys_action'])
del(turn_data['trg_sys_action'])
turn_data['trg_user_action'] = dict()
for domint in da_origin:
domain_intent = da_origin[domint]
_domint = domint.lower()
_domain, _intent = _domint.split('-')
if _domain in cfg.belief_domains:
current_domain = _domain
for slot, p, value in domain_intent:
_slot = slot.lower()
_value = value.strip()
_da = '-'.join((_domint, _slot))
if _da in cfg.da_usr:
turn_data['trg_user_action'][_da] = _value
if _intent == 'inform':
inform_da = _domain+'-'+_slot
if inform_da in cfg.inform_da:
belief_state[_domain][_slot] = _value
if inform_da in cfg.inform_da_usr and _slot in session_data[_domain] \
and session_data[_domain][_slot] != '?':
discard(goal_state[_domain], _slot)
elif _intent == 'request':
request_da = _domain+'-'+_slot
if request_da in cfg.request_da:
belief_state[_domain][_slot] = '?'
else: # sys
book_status = turn['metadata']
for domain in cfg.belief_domains:
if book_status[domain]['book']['booked']:
entity = book_status[domain]['book']['booked'][0]
if 'booked' in belief_state[domain]:
continue
book_domain = domain
if domain in ['taxi', 'hospital', 'police']:
belief_state[domain]['booked'] = f'{domain}-booked'
elif domain == 'train':
found = db.query(domain, [('trainID', entity['trainID'])])
belief_state[domain]['booked'] = found[0]['ref']
else:
found = db.query(domain, [('name', entity['name'])])
belief_state[domain]['booked'] = found[0]['ref']
turn_data['user_action'] = deepcopy(turn_data['trg_user_action'])
del(turn_data['trg_user_action'])
turn_data['others']['change'] = False
turn_data['trg_sys_action'] = dict()
for domint in da_origin:
domain_intent = da_origin[domint]
_domint = domint.lower()
_domain, _intent = _domint.split('-')
for slot, p, value in domain_intent:
_slot = slot.lower()
_value = value.strip()
_da = '-'.join((_domint, _slot, p))
if _da in cfg.da and current_domain:
if _slot == 'ref':
turn_data['trg_sys_action'][_da] = belief_state[book_domain]['booked']
else:
turn_data['trg_sys_action'][_da] = _value
if _intent in ['inform', 'recommend', 'offerbook', 'offerbooked', 'book']:
inform_da = current_domain+'-'+_slot
if inform_da in cfg.request_da:
discard(belief_state[current_domain], _slot, '?')
if inform_da in cfg.request_da_usr and _slot in session_data[current_domain] \
and session_data[current_domain][_slot] == '?':
goal_state[current_domain][_slot] = _value
elif _intent in ['nooffer', 'nobook']:
# TODO: better transition
for da in turn_data['user_action']:
__domain, __intent, __slot = da.split('-')
if __intent == 'inform' and __domain == current_domain:
discard(belief_state[current_domain], __slot)
turn_data['others']['change'] = True
reload(goal_state, session_data, current_domain)
if i + 1 == len(sess['log']):
turn_data['final_belief_state'] = belief_state
turn_data['final_goal_state'] = goal_state
self.data[part].append(deepcopy(turn_data))
add_domain_mask(self.data)
def _set_default(obj):
if isinstance(obj, set):
return list(obj)
raise TypeError
os.makedirs(data_dir_new)
for part in ['train','valid','test']:
with open(data_dir_new + '/' + part + '.json', 'w') as f:
self.data[part] = json.dumps(self.data[part], default=_set_default)
f.write(self.data[part])
self.data[part] = json.loads(self.data[part])
with open(data_dir_new + '/' + part + '_goal.json', 'w') as f:
self.goal[part] = json.dumps(self.goal[part], default=_set_default)
f.write(self.goal[part])
self.goal[part] = json.loads(self.goal[part])
def debug(main_dict):
import ipdb
ipdb.set_trace() # breakpoint 69b4b68e //
metric_name = main_dict["metric_name"]
loss_name = main_dict["loss_name"]
batch_size = main_dict["batch_size"]
mu.print_welcome(main_dict)
train_set, val_set = mu.load_trainval(main_dict)
model, opt, _ = mu.init_model_and_opt(main_dict)
print("Model from scratch...")
batch = ut.get_batch(train_set, indices=[15])
probs = model.predict(batch, "probs")
blobs = model.predict(batch, "blobs")
tr.fitBatch(model, batch, loss_name=loss_name, opt=opt, epochs=100)
val.valBatch(model, batch)
import ipdb
ipdb.set_trace() # breakpoint 5cd16f8f //
ul.visSp_prob(model, batch)
vis.images(batch["images"], aa, denorm=1)
vis.visBlobs(model, batch)
ul.vis_nei(model, batch, topk=1000, thresh=0.8, bg=True)
ul.vis_nei(model, batch, topk=1000, bg=False)
tr.fitQuick(model,
train_set,
batch_size=batch_size,
loss_name=loss_name,
metric_name=metric_name)
val.validate(model,
val_set,
metric_name=main_dict["metric_name"],
batch_size=main_dict["val_batchsize"])
tr.fitQuick(model,
train_set,
batch_size=batch_size,
loss_name=loss_name,
metric_name=metric_name)
tr.fitBatch(model, batch, loss_name=loss_name, opt=opt, epochs=100)
val.valBatch(model, batch_train, metric_name=metric_name)
tr.fitBatch(model,
batch,
loss_function=losses.expand_loss,
opt=opt,
epochs=100)
vis.visBlobs(model, batch)
vis.visWater(model, batch)
val.validate(model, val_set, metric_name="MUCov")
import ipdb
ipdb.set_trace() # breakpoint ddad840d //
model, opt, _ = mu.init_model_and_opt(main_dict)
tr.fitBatch(model, batch, loss_name="water_loss_B", opt=opt, epochs=100)
tr.fitQuick(model, train_set, loss_name=loss_name, metric_name=metric_name)
# vis.images(batch["images"], batch["labels"], denorm=1)
# mu.init.LOSS_DICT["water_loss"](model, batch)
import ipdb
ipdb.set_trace() # breakpoint f304b83a //
vis.images(batch["images"], model.predict(batch, "labels"), denorm=1)
val.valBatch(model, batch, metric_name=main_dict["metric_name"])
vis.visBlobs(model, batch)
import ipdb
ipdb.set_trace() # breakpoint 074c3921 //
tr.fitBatch(model,
batch,
loss_name=main_dict["loss_name"],
opt=opt,
epochs=100)
for e in range(10):
if e == 0:
scoreList = []
scoreList += [
tr.fitIndices(model,
train_set,
loss_name=main_dict["loss_name"],
batch_size=batch_size,
metric_name=metric_name,
opt=opt,
epoch=e,
num_workers=1,
ind=np.random.randint(0, len(train_set), 32))
]
tr.fitData(model, train_set, opt=opt, epochs=10)
ut.reload(sp)
water = sp.watersplit(model, batch).astype(int)
vis.images(batch["images"], water, denorm=1)
vis.visBlobs(model, batch)
vis.images(batch["images"], ul.split_crf(model, batch), denorm=1)
losses.dense_crf(model, batch, alpha=61, beta=31, gamma=1)
vis.visBlobs(model, batch)
model.blob_mode = "superpixels"
#----------------------
# Vis Blobs
vis.visBlobs(model, batch)
vis.images(batch["images"], model.predict(batch, "labels"), denorm=1)
# Vis Blobs
#vis.visBlobs(model, batch)
vis.images(batch["images"],
sp.watersplit_test(model, batch).astype(int),
denorm=1)
#=sp.watersplit(model, batch).astype(int);
# Vis CRF
vis.images(batch["images"],
ul.dense_crf(model,
batch,
alpha=5,
gamma=5,
beta=5,
smooth=False),
denorm=1)
vis.images(batch["images"], ul.dense_crf(model, batch), denorm=1)
# Eval
val.valBatch(model, batch, metric_name=main_dict["metric_name"])
import ipdb
ipdb.set_trace() # breakpoint e9cd4eb0 //
model = mu.load_best_model(main_dict)
val.valBatch(model, batch, metric_name=main_dict["metric_name"])
tr.fitBatch(model, batch, loss_name=main_dict["loss_name"], opt=opt)
vis.visBlobs(model, batch)
import ipdb
ipdb.set_trace() # breakpoint 2167961a //
batch = ut.get_batch(train_set, indices=[5])
tr.fitBatch(model, batch, loss_name=main_dict["loss_name"], opt=opt)
vis.images(batch["images"], model.predict(batch, "probs"), denorm=1)
vis.visBlobs(model, batch)
val.validate(model, val_set, metric_name=main_dict["metric_name"])
val.validate(model, val_set, metric_name="SBD")
def update_belief_sys(self, old_s, a):
"""
update belief/goal state with sys action
"""
s = deepcopy(old_s)
a_index = torch.nonzero(a) # get multiple da indices
self.time_step += 1
s['others']['turn'] = self.time_step
# update sys/user dialog act
s['sys_action'] = dict()
# update belief part
das = [self.cfg.idx2da[idx.item()] for idx in a_index]
das = [da.split('-') for da in das]
sorted(das, key=lambda x: x[0]) # sort by domain
entities = [] if self.topic == '' else self.get_entities(s, self.topic)
return_flag = False
for domain, intent, slot, p in das:
if domain in self.cfg.belief_domains and domain != self.topic:
self.topic = domain
entities = self.get_entities(s, domain)
da = '-'.join((domain, intent, slot, p))
if intent == 'request':
s['sys_action'][da] = '?'
elif intent in ['nooffer', 'nobook'] and self.topic != '':
return_flag = True
if slot in s['belief_state'][self.topic] and s['belief_state'][
self.topic][slot] != '?':
s['sys_action'][da] = s['belief_state'][self.topic][slot]
else:
s['sys_action'][da] = 'none'
elif slot == 'choice':
s['sys_action'][da] = str(len(entities))
elif slot == 'none':
s['sys_action'][da] = 'none'
else:
num = int(p) - 1
if self.topic and len(
entities) > num and slot in self.cfg.mapping[
self.topic]:
typ = self.cfg.mapping[self.topic][slot]
if typ in entities[num]:
s['sys_action'][da] = entities[num][typ]
else:
s['sys_action'][da] = 'none'
else:
s['sys_action'][da] = 'none'
if not self.topic:
continue
if intent in [
'inform', 'recommend', 'offerbook', 'offerbooked',
'book'
]:
discard(s['belief_state'][self.topic], slot, '?')
if slot in s['user_goal'][self.topic] and s['user_goal'][
self.topic][slot] == '?':
s['goal_state'][self.topic][slot] = s['sys_action'][da]
# booked
if intent == 'inform' and slot == 'car': # taxi
if 'booked' not in s['belief_state']['taxi']:
s['belief_state']['taxi']['booked'] = 'taxi-booked'
elif intent in ['offerbooked', 'book'
] and slot == 'ref': # train
if self.topic in ['taxi', 'hospital', 'police']:
s['belief_state'][
self.topic]['booked'] = f'{self.topic}-booked'
s['sys_action'][da] = f'{self.topic}-booked'
elif entities:
book_domain = entities[0]['ref'].split('-')[0]
if 'booked' not in s['belief_state'][
book_domain] and entities:
s['belief_state'][book_domain][
'booked'] = entities[0]['ref']
s['sys_action'][da] = entities[0]['ref']
if return_flag:
for da in s['user_action']:
d_usr, i_usr, s_usr = da.split('-')
if i_usr == 'inform' and d_usr == self.topic:
discard(s['belief_state'][d_usr], s_usr)
reload(s['goal_state'], s['user_goal'], self.topic)
if not self.lock_evalutor:
self.evaluator.add_sys_da(s['sys_action'])
return s
from django.db import connection
from utils import reload
if "keyvalue_store" in connection.introspection.table_names():
reload()
""" Author: Pratyush Kumar, [email protected] Script to train neural network models using tensorflow's implementation of keras """ import utils utils.reload() utils.set_path() # Load some tensorflow utilities import tensorflow as tf from tensorflow.keras.models import Model from tensorflow.keras.layers import Dense, Activation, Dropout, Input, LSTM, Reshape, Lambda, RepeatVector from tensorflow.keras.optimizers import Adam # Load some more python modules import numpy as np import matplotlib.pyplot as plt import math """ One-hot encoding of ToD and DoW """ #sim_data = utils.load("output_data.pkl") qother = utils.load("input_data.pkl") #start_time = input_data[-2] #end_time = input_data[-1]
def _build_data(self, data_dir, data_dir_new, cfg, db):
"""
按照train，dev，test构建session数据
belief_state, goal_state,others,sys_action,trg_sys_action,usr_action,trg_usr_action
以及附加的状态包括
final_belief_state, final_goal_state, next_avaliable_domain, invisible_domains
"""
data_filename = data_dir + '/' + cfg.data_file
with open(data_filename, 'r') as f:
origin_data = json.load(f)
for part in ['train', 'valid', 'test']:
self.data[part] = []
self.goal[part] = {}
valList = []
with open(data_dir + '/' + cfg.val_file) as f:
for line in f:
valList.append(line.split('.')[0])
testList = []
with open(data_dir + '/' + cfg.test_file) as f:
for line in f:
testList.append(line.split('.')[0])
num_sess = 0
for k_sess in origin_data:
sess = origin_data[k_sess]
if k_sess in valList:
part = 'valid'
elif k_sess in testList:
part = 'test'
else:
part = 'train'
turn_data, session_data = init_session(k_sess, cfg)
# belief_state
belief_state = turn_data['belief_state']
# goal_state
goal_state = turn_data['goal_state']
init_goal(session_data, goal_state, sess['goal'], cfg)
# 直接跳过多domain场景
if "SNG" not in k_sess:
continue
# 判断如果数据中没有指定domain场景，则不使用该数据w
contain_domain = False
for domain in session_data:
content = session_data[domain]
if domain in cfg.belief_domains and len(content) > 0:
contain_domain = True
break
if contain_domain is False:
continue
num_sess += 1
# goal
# 完整的用户目标， goal 和 goal_state 的差异在于， goal为任务最终的目标， goal_state 表示用户目标在某一时刻的状态
self.goal[part][k_sess] = deepcopy(session_data)
current_domain = ''
book_domain = ''
turn_data['trg_user_action'] = {}
turn_data['trg_sys_action'] = {}
for i, turn in enumerate(sess['log']):
# 注意区分 turn_data为session数据，turn为原始数据
turn_data['others']['turn'] = i
turn_data['others']['terminal'] = i + 2 >= len(sess['log'])
# 记录本轮的动作
da_origin = turn['dialog_act']
expand_da(da_origin)
turn_data['belief_state'] = deepcopy(
belief_state) # from previous turn
turn_data['goal_state'] = deepcopy(goal_state)
if i % 2 == 0: # user
# sys_action为上一轮系统动作，trg_sys_action为本轮系统动作,
# 因此需要将trg_sys_action赋值给sys_action,之后删除trg_sys_action
turn_data['sys_action'] = deepcopy(
turn_data['trg_sys_action'])
del (turn_data['trg_sys_action'])
turn_data['trg_user_action'] = dict()
for domint in da_origin:
domain_intent = da_origin[domint]
_domint = domint.lower()
_domain, _intent = _domint.split('-')
# current_domain 防止出现booking domain的情况
if _domain in cfg.belief_domains:
current_domain = _domain
for slot, p, value in domain_intent:
_slot = slot.lower()
_value = value.strip()
_da = '-'.join((_domint, _slot))
# 符合要求的用户动作，如果需要限制用户动作的范围，可以通过修改cfg.da_user即可
if _da in cfg.da_usr:
turn_data['trg_user_action'][_da] = _value
if _intent == 'inform':
inform_da = _domain + '-' + _slot
# 根据动作，调整belief_state和goal_state
if inform_da in cfg.inform_da:
belief_state[_domain][_slot] = _value
if inform_da in cfg.inform_da_usr and _slot in session_data[_domain] \
and session_data[_domain][_slot] != '?':
discard(goal_state[_domain], _slot)
elif _intent == 'request':
request_da = _domain + '-' + _slot
if request_da in cfg.request_da:
belief_state[_domain][_slot] = '?'
else: # sys
# metadata 记录book的状态
book_status = turn['metadata']
for domain in cfg.belief_domains:
if book_status[domain]['book']['booked']:
entity = book_status[domain]['book']['booked'][0]
# 表示已经完成booked，可以不需要再更新订购状体
if 'booked' in belief_state[domain]:
continue
book_domain = domain
if domain in ['taxi', 'hospital', 'police']:
belief_state[domain][
'booked'] = f'{domain}-booked'
elif domain == 'train':
found = db.query(
domain, [('trainID', entity['trainID'])])
belief_state[domain]['booked'] = found[0][
'ref']
else:
found = db.query(domain,
[('name', entity['name'])])
belief_state[domain]['booked'] = found[0][
'ref']
# 对用户动作做相同的处理
turn_data['user_action'] = deepcopy(
turn_data['trg_user_action'])
del (turn_data['trg_user_action'])
# 保证只会影响到下一个用户动作
turn_data['others']['change'] = False
turn_data['trg_sys_action'] = dict()
for domint in da_origin:
domain_intent = da_origin[domint]
_domint = domint.lower()
_domain, _intent = _domint.split('-')
for slot, p, value in domain_intent:
_slot = slot.lower()
_value = value.strip()
_da = '-'.join((_domint, _slot, p))
if _da in cfg.da and current_domain:
if _slot == 'ref':
turn_data['trg_sys_action'][
_da] = belief_state[book_domain][
'booked']
else:
turn_data['trg_sys_action'][_da] = _value
if _intent in [
'inform', 'recommend', 'offerbook',
'offerbooked', 'book'
]:
inform_da = current_domain + '-' + _slot
if inform_da in cfg.request_da:
discard(belief_state[current_domain],
_slot, '?')
if inform_da in cfg.request_da_usr and _slot in session_data[current_domain] \
and session_data[current_domain][_slot] == '?':
goal_state[current_domain][
_slot] = _value
elif _intent in ['nooffer', 'nobook']:
# TODO: better transition
for da in turn_data['user_action']:
__domain, __intent, __slot = da.split(
'-')
# 如果系统无法提供服务，则删除上一次用户针对本domain提供的inform
if __intent == 'inform' and __domain == current_domain:
discard(
belief_state[current_domain],
__slot)
# 改变change标志位，表示用户目标发生转变
turn_data['others']['change'] = True
reload(goal_state, session_data,
current_domain)
# 最后一轮，登入final状态
if i + 1 == len(sess['log']):
turn_data['final_belief_state'] = belief_state
turn_data['final_goal_state'] = goal_state
self.data[part].append(deepcopy(turn_data))
print("session number: ", num_sess)
add_domain_mask(self.data)
def _set_default(obj):
if isinstance(obj, set):
return list(obj)
raise TypeError
# 实验一
# 按照单个domain划分
os.makedirs(data_dir_new)
for part in ['train', 'valid', 'test']:
with open(data_dir_new + '/' + part + '.json', 'w') as f:
self.data[part] = json.dumps(self.data[part],
default=_set_default)
f.write(self.data[part])
self.data[part] = json.loads(self.data[part])
with open(data_dir_new + '/' + part + '_goal.json', 'w') as f:
self.goal[part] = json.dumps(self.goal[part],
default=_set_default)
f.write(self.goal[part])
self.goal[part] = json.loads(self.goal[part])
import sys
import pexpect
missing_modules = []
try:
from colour import Color
except:
missing_modules.append('colour')
from mycroft.messagebus.message import Message
try:
reload(mycroftbss)
except NameError:
try:
import mycroftbss # https://github.com/skeledrew/mcbss
except Exception:
missing_modules.append('mycroft brain skill services')
import utils
state = None
class State():
def __init__(self):
def command():
text = request.form.get('text')
if not text:
return '', 400
args = text.split()
# Check if we know about this command
if args[0] == 'add':
success, m = add_movie(' '.join(args[1:]))
if success:
m = m.get_details()
m['text'] = "Added movie:"
return Response(json.dumps(m), mimetype='application/json')
elif args[0] == 'choose':
if len(args) == 1:
rand = Movie.select().order_by(fn.Random())
else:
genre = ' '.join(args[1:])
rand = movies_in_genre(genre).order_by(fn.Random())
try:
m = rand.get().get_details()
m['text'] = random.choice(config.SAYINGS)
return Response(json.dumps(m), mimetype='application/json')
except Movie.DoesNotExist:
return "No movies yet!"
elif args[0] == 'watched':
name = ' '.join(args[1:])
try:
movie = Movie.get(fn.Lower(Movie.name) == name.lower())
except Movie.DoesNotExist:
return "Sorry, I couldn't find that movie. You need to be exact."
movie.watched = True
movie.save()
return "Marked _{}_ as watched".format(movie.name)
elif args[0] == 'unwatch':
# Undocumented commands
name = ' '.join(args[1:])
try:
movie = Movie.get(fn.Lower(Movie.name) == name.lower())
except Movie.DoesNotExist:
return "Sorry, I couldn't find that movie. You need to be exact."
movie.watched = False
movie.save()
return "Marked _{}_ as un-watched".format(movie.name)
elif args[0] == 'list':
data = {'text': "Must specify `movies` or `genres`."}
if len(args) > 1:
if args[1] == 'movies':
if len(args) > 2:
data = format_movies(' '.join(args[2:]))
else:
data = format_movies()
elif args[1] == 'genres':
data = format_genres()
return Response(json.dumps(data), mimetype='application/json')
# Management commands
elif args[0] == 'refresh_genres':
get_genres()
return 'Done'
elif args[0] == 'update':
update()
return '', 204
elif args[0] == 'reload':
reload()
return '', 204
return Response(json.dumps(help_text()), mimetype='application/json')