def p_numexpressions(p):
"""numexpressions : LBRACKET numexpression RBRACKET numexpressions
| LBRACKET numexpression RBRACKET
"""
if len(p) == 5:
p[0] = Tree("array", p[2], p[4])
else:
p[0] = Tree("array", p[2])
global_tree.append((tuple(['numexpressions'] + p[1:])))
def p_unaryiter(p):
"""unaryiter : unaryop unaryexpr unaryiter
| epsilon
"""
if len(p) == 4:
if p[3] is None:
p[0] = Tree(p[1], None, p[2])
else:
p[3].left = p[2]
p[0] = Tree(p[1], None, p[3])
global_tree.append((tuple(['unaryiter'] + p[1:])))
def p_signedterms(p):
"""signedterms : signal term signedterms
| epsilon
"""
if len(p) == 4:
if p[3] is None:
p[0] = Tree(p[1], None, p[2])
else:
p[3].left = p[2]
p[0] = Tree(p[1], None, p[3])
global_tree.append((tuple(['signedterms'] + p[1:])))
def __getitem__(self, idx):
img = Image.open(
os.path.join(self.img_dir,
str(self.partition[idx]) + '.png'))
tree = ET.parse(
os.path.join(self.tree_dir,
str(self.partition[idx]) + '.xml'))
et_root = tree.getroot()
def build_tree(ele, et_ele):
for et_child in et_ele:
new_child = Tree(et_child.tag)
ele.add_child(new_child)
build_tree(new_child, et_child)
ele.add_child(Tree('end'))
root = Tree(et_root.tag)
build_tree(root, et_root)
if self.tree_transform != None:
root = self.tree_transform(root)
if self.img_transform != None:
img = self.img_transform(img)
return {'img': img, 'tree': root}
def __init__(self, time_not_looping=0.05):
self._not_looping = time_not_looping
self._subscriptions = Tree(config.MQTT_HOME, ["Functions"])
self.payload_on = ("ON", True, "True")
self.payload_off = ("OFF", False, "False")
self._retained = []
self.mqtt_home = config.MQTT_HOME
super().__init__()
self.id = "SmartServer"
self.enable_logger(log)
self.username_pw_set(config.MQTT_USER, config.MQTT_PASSWORD)
self.on_connect = self._connected
self.on_message = self._execute_sync
self.on_disconnect = self._on_disconnect
self.connect(config.MQTT_HOST, 1883, 60)
asyncio.ensure_future(self._keep_connected())
def p_unaryexpr(p):
"""unaryexpr : signal factor
| factor
"""
if len(p) == 2:
p[0] = p[1]
else:
p[0] = Tree(p[1], p[2])
global_tree.append((tuple(['unaryexpr'] + p[1:])))
def p_funclist(p):
"""funclist : funcdef funclist
| funcdef
"""
if len(p) == 3:
p[0] = Tree('funclist', p[1], p[2])
else:
p[0] = p[1]
global_tree.append(('funclist', p[1:]))
def forward(self, img, path):
output = []
for i in range(len(path)):
# if path[i] = root
if path[i].parent == None:
img_features = self.cnn(img)
child_c = self.img_to_c(img_features)
child_h = path[0].value.detach().new(1, self.mem_dim).fill_(0.).requires_grad_()
dummy = Tree('')
dummy.state = child_c, child_h
feed_list = [dummy]
else:
feed_list = [path[i].parent]
idx = path[i].parent.children.index(path[i])
for j in range(idx):
child = path[i].parent.children[j]
child_c = child.value.detach().new(1, self.mem_dim).fill_(0.).requires_grad_()
child_h = child.value.detach().new(1, self.mem_dim).fill_(0.).requires_grad_()
child.state = self.tree_lstm(child.value, child_c, child_h)
feed_list.append(child)
# if this is the end
if i == len(path)-1:
idx = len(path[i].children)
else:
idx = path[i].children.index(path[i+1])
for j in range(idx):
child = path[i].children[j]
child_c = child.value.detach().new(1, self.mem_dim).fill_(0.).requires_grad_()
child_h = child.value.detach().new(1, self.mem_dim).fill_(0.).requires_grad_()
child.state = self.tree_lstm(child.value, child_c, child_h)
feed_list.append(child)
child_c, child_h = zip(* map(lambda x: x.state, feed_list))
child_c, child_h = torch.cat(child_c, dim=0), torch.cat(child_h, dim=0)
path[i].state = self.tree_lstm(path[i].value, child_c, child_h)
output.append(self.h_to_word(path[i].state[1]))
return torch.cat(output, dim=0)
def p_lvalue(p):
"""lvalue : IDENT
| IDENT numexpressions
"""
if find_var(p[1], Scope.actual_scope) is None:
semantic_error("variable \'{}\' not declared.".format(p[1]), p)
if len(p) == 2:
p[0] = p[1]
else:
p[0] = Tree(p[1], p[2])
global_tree.append((tuple(['lvalue'] + p[1:])))
def generate_random_tree(size):
T = Tree()
T.add_root(random.randint(0, MAX_VAL))
frontier = deque()
frontier.append(T.root())
while len(T) < size:
p = frontier.popleft()
num_children = random.randint(1, min(4, size - len(T)))
for _ in range(num_children):
T.add_child(p, random.randint(0, MAX_VAL))
frontier.extend(T.children(p))
return T
def predict_tree(model, img, word_dict, tree, max_child=7):
def str_to_tensor(word):
return torch.from_numpy(word_dict[word]).float().to(device)
def get_path(node, root):
path = []
n = node
path.insert(0, n)
while n!=root:
n = n.parent
path.insert(0, n)
return path
def tree_str(node):
out_str = ' '*node.height() + node.str
for child in node.children:
child_str = tree_str(child)
out_str += '\n' + child_str
return out_str
root = Tree(str_to_tensor('root'))
root.str = 'root'
target_queue = [tree]
target_index = 0
queue = [root]
while len(queue) != 0:
parent = queue[0]
path = get_path(parent, root)
pred = model(img, path).detach()
pred_value = pred[-1]
pred_node = Tree(pred_value)
word = list(word_dict.keys())[torch.argmax(pred_node.value)]
# check ans
target_parent = target_queue[0]
if word!=target_parent.children[target_index].value:
print('wrong!!! target:{} predict:{}'.format(target_parent.children[target_index].value, word))
# print(tree_str(root))
# print("====================")
word = target_parent.children[target_index].value
pred_node.str = word
parent.add_child(pred_node)
if pred_node.str == 'end':
target_index = 0
target_queue.pop(0)
queue.pop(0)
else:
target_queue.append(target_parent.children[target_index])
queue.append(pred_node)
target_index += 1
def _decode(self, code):
""" Given a binary code build a tree corresponding to that code.
@param code (int): A 2b-bit number encoding a tree of size b.
@return tree (Tree): A Tree object corresponding to the binary encoding.
"""
binary_code = [int(bit) for bit in "{0:b}".format(code)]
binary_code = binary_code[1:]
# Assert the binary code is a valid encoding of a tree.
zeros = len([x for x in binary_code if x == 0])
ones = len([x for x in binary_code if x == 1])
assert (zeros == ones)
tree = Tree()
elem = 0
cursor = tree.add_root(elem)
for bit in binary_code:
if bit == 0:
elem += 1
cursor = tree.add_child(cursor, elem)
elif bit == 1:
cursor = tree.parent(cursor)
return tree
def p_binaryoperator(p):
"""binaryoperator : numexpression relationaloperator
"""
p[0] = Tree(p[2], p[1])
global_tree.append((tuple(['binaryoperator'] + p[1:])))
class MQTTHandler(MQTTClient):
def __init__(self, time_not_looping=0.05):
self._not_looping = time_not_looping
self._subscriptions = Tree(config.MQTT_HOME, ["Functions"])
self.payload_on = ("ON", True, "True")
self.payload_off = ("OFF", False, "False")
self._retained = []
self.mqtt_home = config.MQTT_HOME
super().__init__()
self.id = "SmartServer"
self.enable_logger(log)
self.username_pw_set(config.MQTT_USER, config.MQTT_PASSWORD)
self.on_connect = self._connected
self.on_message = self._execute_sync
self.on_disconnect = self._on_disconnect
self.connect(config.MQTT_HOST, 1883, 60)
asyncio.ensure_future(self._keep_connected())
def _connected(self, client, userdata, flags, rc):
log.info("Connection returned result: " + connack_string(rc))
self._publishDeviceStats()
self._subscribeTopics()
def _on_disconnect(self, client, userdata, rc):
if rc != 0:
log.warn("Unexpected disconnection.")
async def _keep_connected(self):
log.info("Keeping connected")
while True:
self.loop(0.05)
await asyncio.sleep(self._not_looping)
def _subscribeTopics(self):
for obj, topic in self._subscriptions.__iter__(with_path=True):
super().subscribe(topic, qos=1)
def unsubscribe(self, topic, callback=None):
if self._isDeviceTopic(topic):
topic = self.getRealTopic(topic)
if callback is None:
log.debug("unsubscribing topic {}".format(topic))
self._subscriptions.removeObject(topic)
super().unsubscribe(topic)
else:
try:
cbs = self._subscriptions.getFunctions(topic)
if type(cbs) not in (tuple, list):
self._subscriptions.removeObject(topic)
return
try:
cbs = list(cbs)
cbs.remove(callback)
except ValueError:
log.warn(
"Callback to topic {!s} not subscribed".format(topic),
local_only=True)
return
self._subscriptions.setFunctions(topic, cbs)
except ValueError:
log.warn("Topic {!s} does not exist".format(topic))
async def subscribe(self, topic, callback, qos=0, check_retained=True):
if self._isDeviceTopic(topic):
topic = self.getRealTopic(topic)
log.debug("Subscribing to topic {}".format(topic))
self._subscriptions.addObject(topic, callback)
if check_retained:
if topic[-4:] == "/set":
# subscribe to topic without /set to get retained message for this topic state
# this is done additionally to the retained topic with /set in order to recreate
# the current state and then get new instructions in /set
state_topic = topic[:-4]
self._retained.append(state_topic)
self._subscriptions.addObject(state_topic, callback)
super().subscribe(state_topic, qos)
await self._await_retained(state_topic, callback, True)
# to give retained state time to process before adding /set subscription
self._retained.append(topic)
super().subscribe(topic, qos)
if check_retained:
asyncio.ensure_future(self._await_retained(topic, callback))
def _publishDeviceStats(self):
pass
def getDeviceTopic(self, attrib, is_request=False):
if is_request:
attrib += "/set"
return ".{}".format(attrib)
def _isDeviceTopic(self, topic):
if topic[:1] == ".":
return True
return False
def getRealTopic(self, device_topic):
if device_topic[:1] != ".":
raise ValueError("DeviceTopic does not start with .")
return "{}/{}/{}".format(self.mqtt_home, self.id, device_topic[1:])
async def _await_retained(self, topic, cb=None, remove_after=False):
st = 0
while topic in self._retained and st <= 8:
await asyncio.sleep(0.1)
st += 1
try:
log.debug("removing retained topic {}".format(topic))
self._retained.remove(topic)
except ValueError:
pass
if remove_after:
self.unsubscribe(topic, cb)
def _execute_sync(self, client, userdata, msg):
"""mqtt library only handles sync callbacks so add it to async loop"""
asyncio.ensure_future(self._execute(msg.topic, msg.payload,
msg.retain))
async def _execute(self, topic, msg, retain):
log.debug("mqtt execution: {!s} {!s}".format(topic, msg))
msg = msg.decode()
try:
msg = json.loads(msg)
except:
pass # maybe not a json string, no way of knowing
cb = None
if topic in self._retained:
retain = True
else:
for topicR in self._retained:
if topicR[-1:] == "#":
if topic.find(topicR[:-1]) != -1:
retain = True
if retain:
try:
cb = self._subscriptions.getFunctions(topic + "/set")
except IndexError:
try:
cb = self._subscriptions.getFunctions(topic)
except IndexError:
pass
if cb is None:
try:
cb = self._subscriptions.getFunctions(topic)
except IndexError:
log.warn("No cb found for topic {!s}".format(topic))
if cb:
for callback in cb if (type(cb) == list
or type(cb) == tuple) else [cb]:
try:
if asyncio.iscoroutinefunction(callback):
res = await callback(topic=topic,
msg=msg,
retain=retain)
else:
res = callback(topic=topic, msg=msg, retain=retain)
if not retain:
if (type(res) == int
and res is not None) or res == True:
# so that an integer 0 is interpreted as a result to send back
if res == True and type(res) != int:
res = msg
# send original msg back
if topic[-4:] == "/set":
# if a /set topic is found, send without /set
self.publish(topic[:-4], res, retain=True)
except Exception as e:
log.error(
"Error executing {!s} mqtt topic {!r}: {!s}".format(
"retained " if retain else "", topic, e))
if retain and topic[-2:] != "/#":
# only remove if it is not a wildcard topic to allow other topics
# to handle retained messages belonging to this wildcard
try:
self._retained.remove(topic)
except ValueError:
pass
# already removed by _await_retained
def publish(self, topic, msg, retain=False, qos=0):
if type(msg) == dict or type(msg) == list:
msg = json.dumps(msg)
elif type(msg) != str:
msg = str(msg)
if self._isDeviceTopic(topic):
topic = self.getRealTopic(topic)
super().publish(topic, msg, retain=retain, qos=qos)
def build_tree(ele, et_ele):
for et_child in et_ele:
new_child = Tree(et_child.tag)
ele.add_child(new_child)
build_tree(new_child, et_child)
ele.add_child(Tree('end'))
def predict_tree(img, model, device, word_dict, max_child=6):
tranf = transforms.Compose([WordEmbedding(word_dict), TreeToTensor()])
end_value = tranf(Tree('end')).value.to(device)
model.to(device)
root = Tree('root')
root = tranf(root)
root.value = root.value.to(device)
out_size = root.value.size()
queue = [root]
while len(queue) != 0:
sub_tree = Tree(image_caption_model(img, [root]).flatten().detach())
max_value = torch.max(sub_tree.value)
sub_tree.value = torch.where(sub_tree.value >= max_value,
torch.ones(out_size).to(device),
torch.zeros(out_size).to(device))
queue[0].add_child(sub_tree)
if len(queue[0].children) >= max_child:
sub_tree = Tree(end_value.clone().detach())
queue[0].add_child(sub_tree)
if torch.equal(end_value, sub_tree.value):
queue.pop(0)
else:
queue.append(sub_tree)
root.for_each_value(lambda x: x.cpu().numpy())
vec2word = Vec2Word(word_dict)
root = vec2word(root)
return root
