def test_convert_us(self):
us = [
Unit(1, 10, KEY_PRESS_UNIT_TYPE),
Unit(2, 11, KEY_PRESS_UNIT_TYPE),
Unit(3, 10, KEY_RELEASE_UNIT_TYPE),
Unit(4, 11, KEY_RELEASE_UNIT_TYPE),
]
self.assertEqual(get_inputs_sequence(us, False),
[Input(10, 1, 3), Input(11, 2, 4)])
def main(argv=sys.argv): f = Face() n = Net() o = Output() if(len(argv) >= 2): argv.pop(0) for arg in argv: print(arg) d = Input(arg) o.outFaces(f.detect(d.getImage(Input.FACES))) dogs,cats = n.result(d.getImage(Input.PREDICTION)) o.outAnimals(dogs,cats)
def test_get_i_d(self):
inputs = [
Input(10, 1, 3),
Input(11, 2.1, 4)
]
self.assertEqual(
get_inner_distance_sequence(inputs),
[
InnerDistance(10, 2),
InnerDistance(11, 1.9)
]
)
def test_get_dist(self):
inputs = [
Input(10, 1, 3),
Input(11, 2, 4),
Input(12, 6, 8)
]
self.assertEqual(
get_outer_distance_sequence(inputs),
[
OuterDistance(10, 11, -1),
OuterDistance(11, 12, 2)
]
)
def run():
model = Input(4)
model = Affine(model, 128)
model = LReLU(model)
model = Affine(model, 2)
model = Softmax(model)
world = StochasticPolicy(Gym(make_env, max_steps=500))
opt = Adam(np.random.randn(model.n_params) * 0.1, lr=0.01)
for _ in range(50):
model.load_params(opt.get_value())
trajs = world.trajectories(model, 16)
print_reward(trajs, max_value=5000)
trajs = discount(trajs, horizon=500)
trajs = normalize(trajs)
grad = policy_gradient(trajs, policy=model)
opt.apply_gradient(grad)
while True:
world.render(model)
def get_inputs_sequence_from_messages_sequence(messages_sequence):
input_sequence = list()
for message in messages_sequence:
temp_seq = list()
i = 0
while i < len(message):
if message[i].type == KEY_RELEASE_UNIT_TYPE:
i += 1
continue
current_unit = message[i]
j = 1
while i + j < len(message):
if message[i +
j].key_code == current_unit.key_code and message[
i + j].type == KEY_RELEASE_UNIT_TYPE:
temp_seq.append(
Input(current_unit.key_code, current_unit.time_stamp,
message[i + j].time_stamp))
break
else:
j += 1
i += 1
input_sequence.append(tuple(temp_seq))
return input_sequence
def walker():
walker = Input(STATE_SIZE)
walker = Affine(walker, 64)
walker = LReLU(walker)
walker = Affine(walker, ACTION_SIZE)
walker = Tanh(walker)
return walker
def makeTransaction(node_send, node_rec, val):
get_val = val
transactions = copy.deepcopy(node_send.unSpentTransactions)
inputs = []
for hash in transactions:
tr = transactions[hash]
for index in range(len(tr.outputs)):
output = tr.outputs[index]
if output is not None:
if output.pub_key == node_send.pub_key.to_string():
signature = node_send.priv_key.sign(hash.encode('utf-8'))
input = Input(hash, index, signature)
inputs.append(input)
get_val -= output.value
if get_val < 0.0:
break
if get_val > 0.0:
print("Not enough Coin to make Transaction...")
return False
outputs = []
main_output = Output(node_rec.pub_key.to_string(), val)
outputs.append(main_output)
if get_val < 0.0:
outputs.append(Output(node_send.pub_key.to_string(), 0.0 - get_val))
new_transaction = Transaction(inputs, outputs)
node_send.sendTransaction(new_transaction)
return True
def get_inputs_sequence(unit_sequence, is_raw):
if is_raw:
unit_sequence = remove_intermediate_pressing(
remove_duplicates(unit_sequence))
input_sequence = list()
i = 0
while i < len(unit_sequence):
if unit_sequence[i].type == KEY_RELEASE_UNIT_TYPE:
i += 1
continue
current_unit = unit_sequence[i]
j = 1
while i + j < len(unit_sequence):
if unit_sequence[
i + j].key_code == current_unit.key_code and unit_sequence[
i + j].type == KEY_RELEASE_UNIT_TYPE:
input_sequence.append(
Input(current_unit.key_code, current_unit.time_stamp,
unit_sequence[i + j].time_stamp))
break
else:
j += 1
i += 1
return input_sequence
def build_classifier():
model = Input(2)
model = LReLU(Affine(model, 32))
model = LReLU(Affine(model, 32))
model = Affine(model, 2)
model = Softmax(model)
return model
def createRewardTransaction(node):
hash = 'BLOCK-REWARD'
index = -1
signature = "YOLO"
input = Input(hash, index, signature)
output = Output(node.pub_key.to_string(), 5.0)
transaction = Transaction([input], [output])
print("Transaction: {} being sent...".format(transaction.hash))
node.sendTransaction(transaction)
def run():
classifier = Input(2)
classifier = Affine(classifier, 16)
classifier = LReLU(classifier)
classifier = Affine(classifier, 2)
classifier = Softmax(classifier)
gausses = [Constant(2)]
gausses[0].load_params([0., 0.])
plt.ion()
def train_one():
gaussOpt = Adam(
[0., 0.],
lr=0.010,
memory=0.5,
)
classOpt = Adam(np.random.randn(classifier.n_params) * 0.1,
lr=0.5,
memory=0.99)
gaussCenterer = Constant(2)
gausses.append(gaussCenterer)
curAccuracy = 0.
while curAccuracy < 0.98:
classifier.load_params(classOpt.get_value())
gaussCenterer.load_params(gaussOpt.get_value())
trajs = [[(gauss_observation(gausses[:-1]), [1, 0], 1.)]
for _ in range(500)]
trajs += [[(gauss_observation(gausses[-1:]), [0, 1], 1.)]
for _ in range(500)]
accTrajs = accuracy(trajs, model=classifier)
print_reward(accTrajs, max_value=1.0)
accs = [traj[0][2] for traj in accTrajs]
curAccuracy = np.mean(accs)
grad = policy_gradient(trajs, policy=classifier)
classOpt.apply_gradient(grad)
trajs2 = learn_from_classifier(classifier, trajs[500:], 1)
trajs2 = normalize(trajs2)
grad2 = policy_gradient(trajs2, policy=gaussCenterer)
gaussOpt.apply_gradient(grad2)
plt.clf()
plt.grid()
plt.gcf().axes[0].set_ylim([-1, 1])
plt.gcf().axes[0].set_xlim([-1, 1])
x, y = zip(*[o for ((o, _, _), ) in trajs[:500]])
plt.scatter(x, y, color="blue")
x, y = zip(*[o for ((o, _, _), ) in trajs[500:]])
plt.scatter(x, y, color="red")
plt.pause(0.01)
for i in range(10):
print("Teaching agent %d." % i)
train_one()
plt.pause(10000000000000.)
def load_data(path):
inputs = []
outputs = []
df = pd.read_csv(path)
# inputs = df.question_text.to_numpy()
# print(inputs)
for _, r in df.iterrows():
t = r['question_text'].strip().lower()
id = r['target']
inputs.append(Input(t, id))
score = r['target']
outputs.append(Output(score))
return inputs, outputs
def posts_post():
""" Add a new post """
data = request.json
# Check that the JSON supplied is valid
# If not we return a 422 Unprocessable Entity
try:
validate(data, post_schema)
except ValidationError as error:
data = {"message": error.message}
return Response(json.dumps(data), 422, mimetype="application/json")
# Add the post to the database
post = Input(title=data["title"], rent=data["rent"])
session.add(post)
session.commit()
# Return a 201 Created, containing the post as JSON and with the
# Location header set to the location of the post
data = json.dumps(post.as_dictionary())
headers = {"Location": url_for("post_get", id=post.id)}
return Response(data, 201, headers=headers,
mimetype="application/json")
def run():
model = Input(28, 28)
model = Affine(model, 128)
model = LReLU(model)
model = Affine(model, 10)
model = Softmax(model)
train_world = StochasticPolicy(Accuracy(Mnist()))
opt = Adams(np.random.randn(model.n_params), lr=0.00002, memory=0.99)
for i in range(600):
model.load_params(opt.get_value())
trajs = train_world.trajectories(model, 128)
print_reward(trajs, max_value=1)
grad = policy_gradient(trajs, policy=model)
opt.apply_gradient(grad)
def run():
world = Bytes(b"aabbaab", max_steps=4, charset=b'abcd')
print("\nConstant model:\n")
model = Constant(4)
model = Softmax(model)
train(world, model)
print("\nLast character:\n")
model = Input(4)
model = Affine(model, 4)
model = LReLU(model)
model = Affine(model, 4)
model = Softmax(model)
train(world, model)
print("\nLast two characters:\n")
model = Input(2, 4)
model = Affine(model, 4)
model = LReLU(model)
model = Affine(model, 4)
model = History(model, length=2)
model = Softmax(model)
train(world, model)
def run():
world = StochasticPolicy(Gym("CartPole-v1"))
model = Input(4)
model = Affine(model, 64)
model = LReLU(model)
model = Affine(model, 2)
model = Softmax(model)
if len(sys.argv) >= 2:
params = np.load(sys.argv[1])
else:
params = train(world, model)
np.save("__cartpole.npy", params)
model.load_params(params)
world.render(model)
def fileToObject(filename):
with open(filename, 'r') as fin:
lines = fin.read().splitlines()
numOfBooks, numOfLibs, deadline = map(int, lines[0].split(' '))
bookScores = list(map(int, lines[1].split(' ')))
libs = []
for i in range(2, len(lines), 2):
if (lines[i] == ''):
continue
_, signUp, booksOutput = map(int, lines[i].split(' '))
books = []
for idx in map(int, lines[i+1].split(' ')):
books.append(Book(idx, bookScores[idx]))
libs.append(Library(len(libs), signUp, booksOutput, books, []))
return Input(numOfBooks, deadline, libs)
def run():
if len(sys.argv) < 2:
print("Usage: imitate.py <file>")
return
with open(sys.argv[1], "r") as f:
data = f.buffer.read()
charset = set(data)
world = Bytes(data, max_steps=100, charset=charset)
print("Charset size: %d" % len(charset))
model = Input(len(charset))
model = LSTM(model)
model = Affine(model, len(charset))
model = Softmax(model)
train(world, model)
for _ in range(10):
world.render(model)
def run():
classifier = Input(2)
classifier = Affine(classifier, 16)
classifier = LReLU(classifier)
classifier = Affine(classifier, 2)
classifier = Softmax(classifier)
curCarr = carr()
curCarr.load_params(np.random.randn(curCarr.n_params))
world = Gym("MountainCarContinuous-v0", max_steps=500)
world = ActionNoise(world, stddev=0.1)
world = Curiosity(world,
classifier=classifier,
history_length=800,
plot=True)
def train_one(carrOpt):
if carrOpt == None:
carrOpt = Adam(
np.random.randn(curCarr.n_params),
lr=0.10,
memory=0.5,
)
nextBreak = 5
for i in range(250):
curCarr.load_params(carrOpt.get_value())
realTrajs, curiosityTrajs = world.trajectories(curCarr, 50)
curScore = np.mean(get_rewards(realTrajs, episode=np.sum)) / 90.
print_reward(realTrajs,
max_value=90.0,
episode=np.sum,
label="Real reward: ")
print_reward(curiosityTrajs,
max_value=1.0,
episode=np.max,
label="Curiosity reward: ")
curCuriosity = np.mean(get_rewards(curiosityTrajs, episode=np.max))
if curCuriosity > 0.98:
if nextBreak == 0:
break
else:
nextBreak -= 1
else:
nextBreak = np.min([nextBreak + 1, 5])
realTrajs = replace_rewards(realTrajs, episode=np.sum)
realTrajs = normalize(realTrajs)
curiosityTrajs = replace_rewards(curiosityTrajs, episode=np.max)
#this is stupid, we should care more(?) if the costs are to high
realWeight = 0.001 + np.max([np.min([curScore, 0.2]), 0.
]) * 0.998 / 0.2
curiosityWeight = 1. - realWeight
print('RWeight: %f, CWeight: %f' % (realWeight, curiosityWeight))
trajs = combine_rewards([realTrajs, curiosityTrajs],
[realWeight, curiosityWeight])
trajs = normalize(trajs)
grad = policy_gradient(trajs, policy=curCarr)
carrOpt.apply_gradient(grad)
if i % 10 == 0:
print("%d episodes in." % i)
world.remember_agent(curCarr)
world.render(curCarr)
if curScore > 0.01:
return carrOpt
else:
return None
theCarOpt = None
for i in range(50):
print("Teaching agent %d." % i)
theCarOpt = train_one(theCarOpt)
def carr():
carr = Input(2)
carr = Affine(carr, 32)
carr = LReLU(carr)
carr = Affine(carr, 3)
return Softmax(carr)
def build_oracle():
model = Input(2)
model = LReLU(Affine(model, 32))
model = LReLU(Affine(model, 32))
model = Affine(model, 2)
return model
def build_agent():
model = Input(2)
model = LReLU(Affine(model, 32))
model = LReLU(Affine(model, 32))
model = Affine(model, 1)
return model
def run():
model = Input(28, 28)
model = Conv2d(model, size=3, channels=8)
model = LReLU(model)
model = Maxpool(model, size=2)
model = Conv2d(model, size=5, channels=16)
model = LReLU(model)
model = Maxpool(model, size=2)
model = Affine(model, 128)
model = LReLU(model)
model = Affine(model, 10)
model = Softmax(model)
if len(sys.argv) >= 2:
params = np.load(sys.argv[1])
else:
params = train(model)
np.save("__mnist.npy", params)
model.load_params(params)
test_world = Mnist(test=True)
trajs = test_world.trajectories(None, 5000)
trajs = accuracy(trajs, model=model, percent=True)
print_reward(trajs, max_value=100, label="Test accuracy:")
def carr():
carr = Input(2)
carr = Affine(carr, 32)
carr = LReLU(carr)
carr = Affine(carr, 1)
return carr
def run():
classifier = Input(7)
classifier = Affine(classifier, 32)
classifier = LReLU(classifier)
classifier = Affine(classifier, 2)
classifier = Softmax(classifier)
agent = walker()
agent.load_params(np.random.randn(agent.n_params) * 1.5)
MAX_TRAIN_TIME = 200
trainTimeLeft = MAX_TRAIN_TIME
curAgentId = -1
curMemoryId = 0
def plot_tagged_trajs(trajs):
nonlocal trainTimeLeft, curAgentId, curMemoryId
COLORS = ["blue", "red"]
plt.clf()
plt.grid()
plt.gcf().axes[0].set_xlim([-1.25, 1.25])
plt.gcf().axes[0].set_ylim([-1.25, 1.25])
plt.suptitle("Episode %d of agent %d, memories: %d" %
(MAX_TRAIN_TIME - trainTimeLeft, curAgentId, curMemoryId))
for traj in trajs:
tag = traj[0][1]
xs, ys = [], []
for state, _, _ in traj:
x = state[2]
y = state[3]
xs.append(x)
ys.append(y)
plt.plot(xs, ys, color=COLORS[np.argmax(tag)], alpha=0.1)
plt.gcf().set_size_inches(10, 8)
plt.gcf().savefig("__step_a%03d_t%03d.png" %
(curAgentId, MAX_TRAIN_TIME - trainTimeLeft),
dpi=100)
world = Gym("BipedalWalker-v2", max_steps=MAX_STEPS)
world = ActionNoise(world, stddev=0.2)
world = Curiosity(world,
classifier=classifier,
history_length=50,
for_classifier=lambda ts: change_obs_space(
ts, changer=interesting_part),
plot=plot_tagged_trajs)
MAX_BOREDOM = 3
boredom = MAX_BOREDOM
MAX_MOTIVATION = 3
motivation = MAX_MOTIVATION
agentOpt = None
lastScores = None
def memorize():
nonlocal boredom, curMemoryId
print("Memorizing %d..." % curMemoryId)
world.remember(agent)
boredom = MAX_BOREDOM
curMemoryId += 1
def save_agent():
np.save(
"__ranger_a%03d_t%03d.npy" %
(curAgentId, MAX_TRAIN_TIME - trainTimeLeft), agentOpt.get_value())
def reset_agent():
nonlocal agentOpt, trainTimeLeft, lastScores, curAgentId, motivation
if agentOpt is not None:
save_agent()
print("Resetting agent %d." % curAgentId)
agentOpt = Adam(
np.random.randn(agent.n_params) * 1.5,
lr=0.05,
memory=0.9,
)
trainTimeLeft = MAX_TRAIN_TIME
lastScores = [-0.4]
curAgentId += 1
motivation = MAX_MOTIVATION
reset_agent()
while True:
agent.load_params(agentOpt.get_value())
realTrajs, curiosityTrajs = world.trajectories(agent, 30)
curScore = np.mean(get_rewards(realTrajs, episode=np.sum)) / 300.
lastScores.append(curScore)
lastScores = lastScores[-10:]
scoreDev = np.std(lastScores)
scoreMean = np.max([np.abs(np.mean(lastScores)), 1.])
curCuriosity = np.mean(get_rewards(curiosityTrajs, episode=np.max))
print_reward(realTrajs,
max_value=300.0,
episode=np.sum,
label="Real reward: ")
print_reward(curiosityTrajs,
max_value=1.0,
episode=np.max,
label="Curiosity reward: ")
if curCuriosity > 0.85:
if boredom == 0:
save_agent()
memorize()
else:
boredom -= 1
else:
boredom = np.min([boredom + 1, MAX_BOREDOM])
if scoreDev / scoreMean < 0.010 or trainTimeLeft < 0:
if motivation == 0:
print("Not really learning.")
save_agent()
motivation = MAX_MOTIVATION
trainTimeLeft = MAX_TRAIN_TIME
if curScore < 0.01:
memorize()
reset_agent()
continue
else:
motivation -= 1
else:
motivation = np.min([motivation + 1, MAX_MOTIVATION])
realTrajs = discount(realTrajs, horizon=200)
realTrajs = normalize(realTrajs)
curiosityTrajs = replace_rewards(curiosityTrajs, episode=np.max)
realWeight = np.min([scoreDev / scoreMean * 10., 0.9])
curiosityWeight = 1. - realWeight
trajs = combine_rewards([realTrajs, curiosityTrajs],
[realWeight, curiosityWeight])
trajs = normalize(trajs)
grad = policy_gradient(trajs, policy=agent)
agentOpt.apply_gradient(grad)
trainTimeLeft -= 1
def run():
classifier = Input(2)
classifier = Affine(classifier, 16)
classifier = LReLU(classifier)
classifier = Affine(classifier, 2)
classifier = Softmax(classifier)
world = Gym("MountainCar-v0")
world = StochasticPolicy(world)
curCarr = carr()
curCarr.load_params(np.random.randn(curCarr.n_params))
oldTrajs = world.trajectories(curCarr, 800)
def train_one(carrOpt):
nonlocal oldTrajs
classOpt = Adam(
np.random.randn(classifier.n_params) * 1.,
lr=0.5,
memory=0.9,
)
if carrOpt == None:
carrOpt = Adam(
np.random.randn(curCarr.n_params),
lr=0.10,
memory=0.5,
)
curScore = 0.
curAccuracy = 0.
for i in range(250):
classifier.load_params(classOpt.get_value())
curCarr.load_params(carrOpt.get_value())
oldTrajIdx = np.random.choice(len(oldTrajs), size=50)
trajs = [oldTrajs[i] for i in oldTrajIdx]
trajs += world.trajectories(curCarr, 50)
trajsForClass = [tag_traj(traj, [1, 0]) for traj in trajs[:50]]
trajsForClass += [tag_traj(traj, [0, 1]) for traj in trajs[50:]]
plot_tagged_trajs(trajsForClass)
accTrajs = accuracy(trajsForClass, model=classifier)
print_reward(accTrajs,
max_value=1.0,
episode=np.mean,
label="Cla reward: ")
curAccuracy = np.mean(get_rewards(accTrajs, episode=np.mean))
if curAccuracy > 1. - i / 500:
break
grad = policy_gradient(trajsForClass, policy=classifier)
classOpt.apply_gradient(grad)
trajs2 = learn_from_classifier(classifier, trajs[50:], 1)
print_reward(trajs2,
max_value=1.0,
episode=np.max,
label="Car reward: ")
curScore = np.mean(get_rewards(trajs2, episode=np.max))
trajs2 = replace_rewards(trajs2, episode=np.max)
trajs2 = normalize(trajs2)
grad2 = policy_gradient(trajs2, policy=curCarr)
carrOpt.apply_gradient(grad2)
if i % 10 == 0:
print("%d episodes in." % i)
oldTrajs += world.trajectories(curCarr, 800)
world.render(curCarr)
if curScore > 0.11:
return carrOpt
else:
return None
theCarOpt = None
for i in range(10):
print("Teaching agent %d." % i)
theCarOpt = train_one(theCarOpt)
def message_text(event):
All_state = json.load(open("db.txt"))
user_input = event.message.text
user_id = event.source.user_id
if user_id not in All_state:
All_state[user_id] = 0
user_state = All_state[user_id]
# print(All_state)
if user_state == 0 and (user_input == "E" or user_input == "e"):
text = "請依照下列格式輸入你的基金組合喔\U0010005E:\n\n" + \
"[基金1統編]:[基金1比重],[基金2統編]:[基金2比重]...\n\n" + \
"範例:\n" + \
"26396604B:0.4,26286281F:0.6"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 1
elif user_state == 0:
text = "" + \
"嗨\U0010005E\U0010005E\U0010005E!\n" + \
"這是一個評估/推薦基金組合的聊天機器人\U0010005E。\n\n" + \
"如果想要評估目前的基金組合，請輸入\"E\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
elif user_state == 1 and (user_input == "Q" or user_input == "q"):
text = "" + \
"嗨\U0010005E\U0010005E\U0010005E!\n" + \
"這是一個評估/推薦基金組合的聊天機器人\U0010005E。\n\n" + \
"如果想要評估目前的基金組合，請輸入\"E\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 0
elif user_state == 1:
try:
user_input_splited = user_input.split(",")
selectFunds = []
selectFund_weights = []
for item in user_input_splited:
temp_fund, temp_weight = item.split(":", 1)
selectFunds.append(temp_fund)
selectFund_weights.append(float(temp_weight))
# print(selectFunds,selectFund_weights)
All_selectFunds = json.load(open("selectFunds.txt"))
All_selectFunds_weights = json.load(open("selectFund_weights.txt"))
# print(All_selectFunds,All_selectFunds_weights)
All_selectFunds[user_id] = selectFunds
All_selectFunds_weights[user_id] = selectFund_weights
json.dump(All_selectFunds, open("selectFunds.txt", "w"))
json.dump(All_selectFunds_weights,
open("selectFund_weights.txt", "w"))
# print(All_selectFunds,All_selectFunds_weights)
# evaluate
extra_ratio = 0
recommend_num = 0
user_input = Input(extra_ratio, selectFunds, selectFund_weights,
recommend_num)
origin, result = recommend(user_input, line_bot_api, user_id)
# Check
check_ration = 0
for item in selectFund_weights:
if item < 0:
check_ration = -1
break
check_ration += item
if check_ration != 1:
message = TextSendMessage(text="基金比重錯誤!請重新輸入!")
line_bot_api.reply_message(event.reply_token,
[message1, message2])
elif origin == None:
message = TextSendMessage(text="基金統編錯誤!請重新輸入!")
line_bot_api.reply_message(event.reply_token,
[message1, message2])
else:
message = TextMessage(text="評估結果:")
message1 = TextSendMessage(text=origin.output_portfolio())
confirm_template_message = TemplateSendMessage(
alt_text='Confirm template',
template=ConfirmTemplate(text='需要推薦基金組合嗎?',
actions=[
MessageAction(label='是',
text='Y'),
MessageAction(label='否',
text='N')
]))
line_bot_api.push_message(user_id, message)
line_bot_api.push_message(user_id, message1)
line_bot_api.reply_message(event.reply_token,
confirm_template_message)
All_state[user_id] = 2
except:
text = "格式錯誤或是基金不存在\U0010005E!請再輸入一次，如果要停止評估，請輸入\"Q\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
elif user_state == 2 and user_input == "N":
text = "" + \
"嗨\U0010005E\U0010005E\U0010005E!\n" + \
"這是一個評估/推薦基金組合的聊天機器人\U0010005E。\n\n" + \
"如果想要評估目前的基金組合，請輸入\"E\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 0
elif user_state == 2:
text = "" + \
"請依照下列格式輸入資料:\n\n" + \
"[轉換金額比例],[最多推薦的基金數量]\n\n" + \
"範例:\n" + \
"0.3,2\n\n" + \
"**注意**\n" + \
"轉換比例不能大於" + str(LIMIT_FUNDS_RATIO) + "\n" + \
"推薦基金數量不能大於" + str(LIMIT_FUNDS_NUM)
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 3
elif user_state == 3 and (user_input == "Q" or user_input == "q"):
text = "" + \
"嗨\U0010005E\U0010005E\U0010005E!\n" + \
"這是一個評估/推薦基金組合的聊天機器人\U0010005E。\n\n" + \
"如果想要評估目前的基金組合，請輸入\"E\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 0
elif user_state == 3:
try:
temp1, temp2 = user_input.split(",", 1)
extra_ratio = float(temp1)
recommend_num = int(temp2)
#check
if extra_ratio > float(LIMIT_FUNDS_RATIO):
message = TextSendMessage(text="轉換金額比例大於限制!請重新輸入!")
line_bot_api.reply_message(event.reply_token, message)
elif recommend_num > int(LIMIT_FUNDS_NUM):
message = TextSendMessage(text="推薦基金數量大於限制!請重新輸入!")
line_bot_api.reply_message(event.reply_token, message)
else:
All_selectFunds = json.load(open("selectFunds.txt"))
All_selectFunds_weights = json.load(
open("selectFund_weights.txt"))
selectFunds = All_selectFunds[user_id]
selectFund_weights = All_selectFunds_weights[user_id]
# print(extra_ratio,recommend_num,selectFunds,selectFund_weights)
#Optimize
user_input = Input(extra_ratio, selectFunds,
selectFund_weights, recommend_num)
origin, result = recommend(user_input, line_bot_api, user_id)
message = TextMessage(text="下列是基金組合的新舊對照\U0010005E:")
message1 = TextSendMessage(text="舊:\n" +
origin.output_portfolio())
message2 = TextSendMessage(text="新:\n" +
result.output_portfolio())
message_last = TextMessage(text="如果還想要評估基金組合，請輸入\"E\"。")
line_bot_api.push_message(user_id, message)
line_bot_api.push_message(user_id, [message1, message2])
line_bot_api.reply_message(event.reply_token, message_last)
All_state[user_id] = 0
except:
text = "格式錯誤\U0010005E!請再輸入一次，如果要停止推薦，請輸入\"Q\"。"
message = TextMessage(text=text)
line_bot_api.reply_message(event.reply_token, message)
else:
message = TextMessage(text="\U0010005E\U0010005E\U0010005E")
line_bot_api.reply_message(event.reply_token, message)
All_state[user_id] = 0
# print(All_state)
json.dump(All_state, open("db.txt", "w"))