def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
price_data = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(price_data[0:10]))
print("price_data idx last 10" + str(price_data[-1]))
input_len = 400
n_actions = 3
obs_size = input_len + n_actions # shape#env.observation_space.shape[0]
training_set = copy.copy(price_data)
X_train = []
y_train = []
for i in range(input_len, len(training_set) - 1001):
# X_train.append(np.flipud(training_set_scaled[i-60:i]))
X_train.append(training_set[i - input_len:i])
y_train.append(training_set[i])
price = y_train
money = 300
before_money = money
ethereum = 0.01
total_money = money + np.float64(price[0] * ethereum)
first_total_money = total_money
pass_count = 0
buy_sell_count = 0 # buy+ sell -
pass_renzoku_count = 0
def create_data():
trade = TradeClass()
# time_date, price_data = trade.getDataPoloniex()
num_history_data = trade.read_bitflyer_json()
X = []
for idx in range(0, len(num_history_data) - INPUT_LEN - SEQUENTIAL_NUM):
tmp = [
num_history_data[idx + list_idx]
for list_idx in range(0, INPUT_LEN)
]
X.append(tmp)
#次の入力データXの最新のデータを、Xの教師Yとする。
Y = []
tmp = []
for i in range(INPUT_LEN, len(num_history_data) - SEQUENTIAL_NUM):
for j in range(0, SEQUENTIAL_NUM):
tmp.append(float(num_history_data[i + j]))
Y.append(tmp)
tmp = []
print(Y[0])
print("len(Y)" + str(len(Y)))
print("len(X))" + str(len(X)))
#XとYの配列の長さが等しくなるように、Xの要素を一つ削除する 。
return X, Y
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
training_set = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(training_set[0:10]))
print("price_data idx last 10" + str(training_set[-1]))
self.input_len = 400
self.n_actions = 3
self.obs_size = self.input_len + self.n_actions #env.observation_space.shape[0]
self.X_train = []
self.y_train = []
for i in range(self.input_len, len(training_set) - 1001):
# X_train.append(np.flipud(training_set_scaled[i-60:i]))
self.X_train.append(training_set[i - self.input_len:i])
self.y_train.append(training_set[i])
self.price = self.y_train
self.money = 300
self.before_money = copy.deepcopy(self.money)
self.cripto = 0.1
self.total_money = self.money + np.float64(self.price[0] * self.cripto)
self.first_total_money = self.total_money
self.pass_count = 0
self.buy_sell_count = 0 # buy+ sell -
self.pass_renzoku_count = 0
self.buy_sell_fee = 0.00001
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
training_set = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(training_set[0:10]))
print("price_data idx last 10" + str(training_set[-1]))
self.length_data = len(training_set)
self.input_len = 400
self.n_actions = 3
self.asset_info_len = 2
self.buy_sell_count_len = 4
#TODO
self.observe_size = self.input_len + self.buy_sell_count_len #+self.asset_info_len #env.observation_space.shape[0]
self.data = []
self.y_train = []
for i in range(self.input_len, self.length_data - 1001):
# data.append(np.flipud(training_set_scaled[i-60:i]))
self.data.append(training_set[i - self.input_len:i])
self.y_train.append(training_set[i])
self.price = self.y_train
self.money = 500
self.before_money = copy.deepcopy(self.money)
self.cripto = 0.00001
self.total_money = self.money + np.float64(self.price[0] * self.cripto)
self.first_total_money = self.total_money
self.pass_count = 0
self.buy_sell_count = 0 # buy+ sell -
self.pass_renzoku_count = 0
self.buy_sell_fee = 0.00001
self.current_asset = [self.cripto, self.money]
self.action_space = gym.spaces.Discrete(3) # 東西南北
self.MAP = np.array([0 for idx in range(0, self.observe_size)])
self.inventory = []
self.observation_space = gym.spaces.Box(low=0,
high=3,
shape=self.MAP.shape)
self.begin_total_money = self.money + self.cripto * self.price[0]
print("LENGTH OF LOOP NUM:" + str(len(self.data)))
self.buy_inventory = []
self.sell_inventory = []
self.total_profit = 0
import numpy as np
import random
import sys, os, copy, traceback
from trade_class import TradeClass
#from trade_class import buy_simple, sell_simple, pass_simple
#from trade_class import SellAndCalcAmoutUsingPrediction,BuyAndCalcAmoutUsingPrediction
cp = cuda.cupy
from sklearn import preprocessing
ss = preprocessing.StandardScaler()
print(os.path.basename(__file__))
tradecl = TradeClass()
price_data = tradecl.ReadPoloniexCSV()
np.set_printoptions(threshold=np.inf)
print("price_data idx 0-10" + str(price_data[0:10]))
print("price_data idx last 10" + str(price_data[-1]))
input_price_len = 400
input_discrete_value_size = 3
total_input_size = input_price_len + input_discrete_value_size
n_actions = 3
# obs_size = input_len+n_actions#shape#env.observation_space.shape[0]
def standarization(x, axis=None):
x = np.array(x)
#coding: utf-8
#https://qiita.com/aaatsushi_bb/items/0b605c0f27493f005c88
import numpy
from trade_class import TradeClass
import xgboost as xgb
import copy
import numpy as np
from sklearn.grid_search import GridSearchCV
import matplotlib as plt
import matplotlib.pyplot as plt
trade = TradeClass()
time_date, price_data = trade.getDataPoloniex()
max_price = max(price_data)
training_set = copy.copy(price_data)
X_train = []
y_train = []
for i in range(60, len(training_set) - 10001):
#X_train.append(np.flipud(training_set_scaled[i-60:i]))
X_train.append(training_set[i - 60:i])
y_train.append(training_set[i])
X_train, y_train = trade.PercentageLabel(X_train, y_train)
X_train, y_train = np.array(X_train,
dtype='float64'), np.array(y_train,
dtype='float64')
from scipy.sparse import coo_matrix
class FxEnv(gym.Env):
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
training_set = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(training_set[0:10]))
print("price_data idx last 10" + str(training_set[-1]))
self.input_len = 400
self.n_actions = 3
self.obs_size = self.input_len + self.n_actions #env.observation_space.shape[0]
self.X_train = []
self.y_train = []
for i in range(self.input_len, len(training_set) - 1001):
# X_train.append(np.flipud(training_set_scaled[i-60:i]))
self.X_train.append(training_set[i - self.input_len:i])
self.y_train.append(training_set[i])
self.price = self.y_train
self.money = 300
self.before_money = copy.deepcopy(self.money)
self.cripto = 0.1
self.total_money = self.money + np.float64(self.price[0] * self.cripto)
self.first_total_money = self.total_money
self.pass_count = 0
self.buy_sell_count = 0 # buy+ sell -
self.pass_renzoku_count = 0
self.buy_sell_fee = 0.00001
def _reset(self):
self.price_idx = 0
return self.X_train[self.price_idx]
def _seed(self, seed=None):
pass
def buy_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = money * 0.1
money -= spend * (1 + self.buy_sell_fee)
if money <= 0.0:
return first_money, first_cripto, first_total_money
cripto += float(spend / current_price)
total_money = money + cripto * current_price
return money, cripto, total_money
def sell_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = cripto * 0.1
cripto -= spend * (1 + self.buy_sell_fee)
if cripto <= 0.0:
return first_money, first_cripto, first_total_money
money += float(spend * current_price)
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def pass_simple(self, money, cripto, total_money, current_price):
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def _step(self, action):
self.price_idx += 1
current_price = self.X_train[self.price_idx][-1]
buy_sell_num_array = [
1.0, 0.0, self.buy_sell_count
] if self.buy_sell_count >= 1 else [0.0, 1.0, self.buy_sell_count]
self.trade.update_trading_view(current_price, action)
pass_reward = 0
if action == 0:
print("buy")
self.buy_sell_count += 1
self.money, self.cripto, self.total_money = self.buy_simple(
self.money, self.cripto, self.total_money, current_price)
elif action == 1:
print("sell")
self.buy_sell_count -= 1
self.money, self.cripto, self.total_money = self.sell_simple(
self.money, self.cripto, self.total_money, current_price)
else:
print("PASS")
self.money, self.cripto, self.total_money = self.pass_simple(
self.money, self.cripto, self.total_money, current_price)
pass_reward = 0.0 # -0.001)#0.01 is default
self.pass_count += 1
reward = self.total_money - self.before_money + pass_reward
if self.buy_sell_count >= 5 and action == 0:
print("buy_sell" + str(self.buy_sell_count) + "回 action==" +
str(action))
reward -= (float(abs(self.buy_sell_count)**2))
print(reward)
elif self.buy_sell_count <= -5 and action == 1:
print("buy_sell" + str(self.buy_sell_count) + "回 action==" +
str(action))
reward -= (float(abs(self.buy_sell_count)**2))
print(reward)
else:
# reward 1.0がちょうど良い！
reward += 1.1
self.before_money = copy.deepcopy(self.total_money)
if self.price_idx % 2000 == 1000:
print("last action:" + str(action))
print("TOTAL MONEY" + str(self.total_money))
print("100回中passは" + str(self.pass_count) + "回")
# print("100回中buy_sell_countは" + str(self.buy_sell_count) + "回")
self.pass_count = 0
self.trade.draw_trading_view()
current_asset = [self.cripto, self.money]
# obs, reward, done, infoを返す
return [
self.X_train[self.price_idx], buy_sell_num_array, current_asset
], reward, False, None
#coding: utf-8
import numpy as np
import matplotlib.pyplot as plt
import time
import datetime
import copy
from trade_class import TradeClass
trade = TradeClass()
time_date, price_data = trade.getDataPoloniex()
max_price = max(price_data)
training_set = copy.copy(price_data)
X_train = []
y_train = []
for i in range(60, len(training_set) - 10001):
#X_train.append(np.flipud(training_set_scaled[i-60:i]))
X_train.append(training_set[i - 60:i])
y_train.append(training_set[i])
X_train, y_train = trade.PercentageLabel(X_train, y_train)
X_train, y_train = np.array(X_train,
dtype='float64'), np.array(y_train,
dtype='float64')
from scipy.sparse import coo_matrix
X_sparse = coo_matrix(X_train)
if bid_ask == "bid":
print("cleaning order Bid Try"+str(counter)+"回目")
bids_top_price = public_zaif.depth('btc_jpy')['bids'][0][0]
limit_price = bids_top_price + counter*5
else:
print("cleaning order Ask Try"+str(counter)+"回目")
asks_top_price = public_zaif.depth('btc_jpy')['asks'][0][0]
limit_price = asks_top_price - counter*5
leverage_zaif.change_position(type="futures", leverage_id=id, group_id=1, price=int(price), limit=int(limit_price))
print("Done. cleaning order")
time.sleep(10)
except:
traceback.print_exc()
print(sys.exc_info()[0])
trade=TradeClass()
history=get_price()
print("History data is here. execute get_price() function")
print(history)
#買いは+、売りは−
leverage_pos_num = -33
for num in range(99999999999):
#time.sleep(60*5)#本当は5分×60セット待たなければならない
try:
time.sleep(20)
public_zaif = ZaifPublicApi()
a = public_zaif.last_price(('btc_jpy'))
print(str(a))
last_price = int(a['last_price'])
class FxEnv(gym.Env):
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
training_set = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(training_set[0:10]))
print("price_data idx last 10" + str(training_set[-1]))
self.length_data = len(training_set)
self.input_len = 400
self.n_actions = 3
self.asset_info_len = 2
self.buy_sell_count_len = 4
#TODO
self.observe_size = self.input_len + self.buy_sell_count_len #+self.asset_info_len #env.observation_space.shape[0]
self.data = []
self.y_train = []
for i in range(self.input_len, self.length_data - 1001):
# data.append(np.flipud(training_set_scaled[i-60:i]))
self.data.append(training_set[i - self.input_len:i])
self.y_train.append(training_set[i])
self.price = self.y_train
self.money = 500
self.before_money = copy.deepcopy(self.money)
self.cripto = 0.00001
self.total_money = self.money + np.float64(self.price[0] * self.cripto)
self.first_total_money = self.total_money
self.pass_count = 0
self.buy_sell_count = 0 # buy+ sell -
self.pass_renzoku_count = 0
self.buy_sell_fee = 0.00001
self.current_asset = [self.cripto, self.money]
self.action_space = gym.spaces.Discrete(3) # 東西南北
self.MAP = np.array([0 for idx in range(0, self.observe_size)])
self.inventory = []
self.observation_space = gym.spaces.Box(low=0,
high=3,
shape=self.MAP.shape)
self.begin_total_money = self.money + self.cripto * self.price[0]
print("LENGTH OF LOOP NUM:" + str(len(self.data)))
self.buy_inventory = []
self.sell_inventory = []
self.total_profit = 0
def _reset(self):
self.price_idx = 0
return self.data[self.price_idx] + [0, 0, 0, 0
] #TODO: +self.current_asset
def _seed(self, seed=None):
return self.length_data
def return_lenghth_steps(self):
return self.length_data
def _render(self, mode='', close=False):
#画面への表示 主にGUI
pass
def buy_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = money * 0.1
money -= spend * (1 + self.buy_sell_fee)
if money <= 0.0:
return first_money, first_cripto, first_total_money
cripto += float(spend / current_price)
total_money = money + cripto * current_price
return money, cripto, total_money
def sell_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = cripto * 0.1
cripto -= spend * (1 + self.buy_sell_fee)
if cripto <= 0.0:
return first_money, first_cripto, first_total_money
money += float(spend * current_price)
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def pass_simple(self, money, cripto, total_money, current_price):
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def buy_lot(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = current_price * 0.001
money -= spend * (1 + self.buy_sell_fee)
EMPTY_MONEY_FLAG = False
if money <= 0.0:
EMPTY_MONEY_FLAG = True
return first_money, first_cripto, first_total_money, EMPTY_MONEY_FLAG
cripto += float(spend / current_price)
total_money = money + cripto * current_price
return money, cripto, total_money, EMPTY_MONEY_FLAG
def sell_lot(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = cripto * 0.001
cripto -= spend * (1 + self.buy_sell_fee)
EMPTY_MONEY_FLAG = False
if cripto <= 0.0:
EMPTY_MONEY_FLAG = True
return first_money, first_cripto, first_total_money, EMPTY_MONEY_FLAG
money += float(spend * current_price)
total_money = money + float(cripto * current_price)
return money, cripto, total_money, EMPTY_MONEY_FLAG
def _step(self, action):
if type(action) is list or type(action) is np.ndarray:
action = action.tolist()
action = action.index(max(action))
else:
pass
self.price_idx += 1
current_price = self.data[self.price_idx][-1]
self.trade.update_trading_view(current_price, action)
len_buy = len(self.buy_inventory)
len_sell = len(self.sell_inventory)
if len_buy > 40:
buy_flag = 1
sell_flag = 0
elif len_sell > 40:
buy_flag = 0
sell_flag = 1
else:
buy_flag = 0
sell_flag = 0
buy_sell_array = [len_buy, len_sell, buy_flag, sell_flag]
#TODO idx + 1じゃなくて良いか？ バグの可能性あり。=>修正済み
#next_state = getStateFromCsvData(self.data, self.price_idx+1, window_size)
reward = 0
if action == 1 and len(self.sell_inventory) > 0: # sell
i = 0
for i in range(0, int(len(self.sell_inventory) / 10)):
sold_price = self.sell_inventory.pop(0)
profit = sold_price - current_price
reward = profit #max(profit, 0)
self.total_profit += profit
#print("Buy(空売りの決済): " + str(current_price) + " | Profit: " + str(profit))
reward = reward / (i + 1)
elif action == 1 and len(self.buy_inventory) < 50: # buy
self.buy_inventory.append(current_price)
#print("Buy: " + str(current_price))
elif action == 2 and len(self.buy_inventory) > 0: # sell
i = 0
for i in range(0, int(len(self.buy_inventory) / 10)):
bought_price = self.buy_inventory.pop(0)
profit = current_price - bought_price
reward = profit # max(profit, 0)
self.total_profit += profit
#print("Sell: " + str(current_price) + " | Profit: " + formatPrice(profit))
reward = reward / (i + 1)
elif action == 2 and len(self.sell_inventory) < 50:
self.sell_inventory.append(current_price)
#print("Sell(空売り): " + formatPrice(current_price))
#print("Reward: "+str(reward))
print("inventory(sell) : " + str(len(self.sell_inventory)) +
" inventory(buy) : " + str(len(self.buy_inventory)))
print("TOTAL PROFIT: " + str(self.total_profit))
if False: #self.price_idx % 10000 == 1000:
try:
print("TOTAL PROFIT: " + str(self.total_profit))
print("inventory(sell) : " + str(len(self.sell_inventory)) +
" inventory(buy) : " +
str(len(self.buy_inventory)))
self.trade.draw_trading_view()
except:
pass
done = True if self.price_idx == self.length_data - 1 else False
# obs, reward, done, infoを返す
return self.data[self.price_idx] + buy_sell_array, reward, done, {}
#from chainerrl.action_value import DiscreteActionValue
#from chainerrl.action_value import QuadraticActionValue
#from chainerrl.optimizers import rmsprop_async
from chainerrl import links
from chainerrl import policies
import numpy as np
import random
import time
import poloniex
import datetime
import copy
from trade_class import TradeClass
trade = TradeClass()
price_data = trade.read_bitflyer_json()
print("price_data idx 0-10" + str(price_data[0:10]))
print("price_data idx last 10" + str(price_data[-1]))
'''
def getDataPoloniex():
polo = poloniex.Poloniex()
polo.timeout = 10
chartUSDT_BTC = polo.returnChartData('USDT_BTC', period=300, start=time.time() - 1440 * 60 * 500, end=time.time()) # 1440(min)*60(sec)=DAY
tmpDate = [chartUSDT_BTC[i]['date'] for i in range(len(chartUSDT_BTC))]
date = [datetime.datetime.fromtimestamp(tmpDate[i]) for i in range(len(tmpDate))]
data = [float(chartUSDT_BTC[i]['open']) for i in range(len(chartUSDT_BTC))]
return date, data
'''
#time_date, price_data = getDataPoloniex()
class FxEnv(gym.Env):
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
training_set = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(training_set[0:10]))
print("price_data idx last 10" + str(training_set[-1]))
self.input_len = 400
self.n_actions = 3
self.asset_info_len = 2
self.buy_sell_count_len = 4
self.observe_size = self.input_len + self.buy_sell_count_len + self.asset_info_len # env.observation_space.shape[0]
self.X_train = []
self.y_train = []
for i in range(self.input_len, len(training_set) - 1001):
# X_train.append(np.flipud(training_set_scaled[i-60:i]))
self.X_train.append(training_set[i - self.input_len:i])
self.y_train.append(training_set[i])
self.price = self.y_train
self.money = 500
self.before_money = copy.deepcopy(self.money)
self.cripto = 0.00001
self.total_money = self.money + np.float64(self.price[0] * self.cripto)
self.first_total_money = self.total_money
self.pass_count = 0
self.buy_sell_count = 0 # buy+ sell -
self.pass_renzoku_count = 0
self.buy_sell_fee = 0.00001
self.current_asset = [self.cripto, self.money]
self.action_space = gym.spaces.Discrete(3) # 東西南北
self.MAP = np.array([0 for idx in range(0, self.observe_size)])
self.inventory = []
self.observation_space = gym.spaces.Box(
low=0,
high=3,
shape=self.MAP.shape
)
self.begin_total_money = self.y_train[0]
print("LENGTH OF LOOP NUM:" + str(len(self.X_train)))
def _reset(self):
self.price_idx = 0
return self.X_train[self.price_idx] + [0, 0, 0, 0] + self.current_asset
def _seed(self, seed=None):
pass
def _render(self, mode='', close=False):
# 画面への表示 主にGUI
pass
def buy_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = money * 0.1
money -= spend * (1 + self.buy_sell_fee)
if money <= 0.0:
return first_money, first_cripto, first_total_money
cripto += float(spend / current_price)
total_money = money + cripto * current_price
return money, cripto, total_money
def sell_simple(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = cripto * 0.1
cripto -= spend * (1 + self.buy_sell_fee)
if cripto <= 0.0:
return first_money, first_cripto, first_total_money
money += float(spend * current_price)
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def pass_simple(self, money, cripto, total_money, current_price):
total_money = money + float(cripto * current_price)
return money, cripto, total_money
def buy_lot(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = current_price * 0.001
money -= spend * (1 + self.buy_sell_fee)
EMPTY_MONEY_FLAG = False
if money <= 0.0:
EMPTY_MONEY_FLAG = True
return first_money, first_cripto, first_total_money, EMPTY_MONEY_FLAG
cripto += float(spend / current_price)
total_money = money + cripto * current_price
return money, cripto, total_money, EMPTY_MONEY_FLAG
def sell_lot(self, money, cripto, total_money, current_price):
first_money, first_cripto, first_total_money = money, cripto, total_money
spend = cripto * 0.001
cripto -= spend * (1 + self.buy_sell_fee)
EMPTY_MONEY_FLAG = False
if cripto <= 0.0:
EMPTY_MONEY_FLAG = True
return first_money, first_cripto, first_total_money, EMPTY_MONEY_FLAG
money += float(spend * current_price)
total_money = money + float(cripto * current_price)
return money, cripto, total_money, EMPTY_MONEY_FLAG
def _step(self, action):
if type(action) is list or type(action) is np.ndarray:
action = action.tolist()
action = action.index(max(action))
else:
pass
self.price_idx += 1
current_price = self.X_train[self.price_idx][-1]
if abs(self.buy_sell_count) >= 10:
between_range = 0.0
else:
between_range = 1.0
buy_sell_num_array = [1.0, 0.0, abs(self.buy_sell_count), between_range] if self.buy_sell_count >= 0 else [0.0,
1.0,
abs(
self.buy_sell_count),
between_range]
self.trade.update_trading_view(current_price, action)
reward = 0
if action == 0:
print("buy")
self.buy_sell_count += 1
self.money, self.cripto, self.total_money = self.buy_simple(self.money, self.cripto, self.total_money,
current_price)
if self.buy_sell_count < 0:
reward += 0.01
else:
reward -= 0.01
elif action == 1 and len(self.inventory) > 0:
print("sell")
self.buy_sell_count -= 1
self.money, self.cripto, self.total_money = self.sell_simple(self.money, self.cripto, self.total_money,
current_price)
if self.buy_sell_count > 0:
reward += 0.01
else:
reward -= 0.01
else:
print("PASS")
self.money, self.cripto, self.total_money = self.pass_simple(self.money, self.cripto, self.total_money,
current_price)
reward += 0.000000 # -0.001)#0.01 is default
self.pass_count += 1
reward += 0.01 * (self.total_money - self.before_money)
print("buy_sell" + str(self.buy_sell_count) + "回 action==" + str(action))
self.before_money = self.total_money
if False: # self.price_idx % 50000 == 1000:
print("last action:" + str(action))
print("TOTAL MONEY" + str(self.total_money))
print("100回中passは" + str(self.pass_count) + "回")
# print("100回中buy_sell_countは" + str(self.buy_sell_count) + "回")
self.pass_count = 0
try:
self.trade.draw_trading_view()
except:
pass
print("begin MONEY: " + str(self.begin_total_money))
print("current MONEY: " + str(self.total_money))
print("price_IDX: " + str(self.price_idx))
print("Reward: " + str(reward))
self.current_asset = [self.cripto, self.money]
# obs, reward, done, infoを返す
return self.X_train[self.price_idx] + buy_sell_num_array + self.current_asset, reward, False, {}
from keras.layers.core import Dense, Activation
from keras.layers.recurrent import LSTM
from keras.optimizers import Adam
from keras.callbacks import EarlyStopping
import numpy as np
import random
import sys, os, copy, traceback
from sklearn.utils import shuffle
from trade_class import TradeClass
from sklearn import preprocessing
ss = preprocessing.StandardScaler()
print(os.path.basename(__file__))
tradecl = TradeClass()
price_data = tradecl.ReadPoloniexCSV()
np.set_printoptions(threshold=np.inf)
print("price_data idx 0-10" + str(price_data[0:10]))
print("price_data idx last 10" + str(price_data[-1]))
input_price_len = 400
input_discrete_value_size = 3
total_input_size = input_price_len + input_discrete_value_size
n_actions = 3
#obs_size = input_len+n_actions#shape#env.observation_space.shape[0]
#データを標準化して、ディープラーニングで学習しやすくする。
def standarization(x, axis=None):
x = np.array(x)
class FxEnv(gym.Env):
def __init__(self):
self.price_idx = 0
self.trade = TradeClass()
price_data = self.trade.read_bitflyer_json()
print("price_data idx 0-10" + str(price_data[0:10]))
print("price_data idx last 10" + str(price_data[-1]))
input_len = 400
n_actions = 3
obs_size = input_len + n_actions # shape#env.observation_space.shape[0]
training_set = copy.copy(price_data)
X_train = []
y_train = []
for i in range(input_len, len(training_set) - 1001):
# X_train.append(np.flipud(training_set_scaled[i-60:i]))
X_train.append(training_set[i - input_len:i])
y_train.append(training_set[i])
price = y_train
money = 300
before_money = money
ethereum = 0.01
total_money = money + np.float64(price[0] * ethereum)
first_total_money = total_money
pass_count = 0
buy_sell_count = 0 # buy+ sell -
pass_renzoku_count = 0
def _reset(self):
self.price_idx = 0
return X_train[0]
def _step(self, action):
self.price_idx += 1
reward = 0
current_price = X_train[self.price_idx][-1]
buy_sell_num_flag = [
1.0, 0.0, buy_sell_count
] if buy_sell_count >= 1 else [0.0, 1.0, buy_sell_count]
self.trade.update_trading_view(current_price, action)
pass_reward = 0
if action == 0:
print("buy")
buy_sell_count += 1
money, ethereum, total_money = buy_simple(money, ethereum,
total_money,
current_price)
elif action == 1:
print("sell")
buy_sell_count -= 1
money, ethereum, total_money = sell_simple(money, ethereum,
total_money,
current_price)
else:
print("PASS")
money, ethereum, total_money = pass_simple(money, ethereum,
total_money,
current_price)
pass_reward = 0.0 # -0.001)#0.01 is default
self.pass_count += 1
reward = total_money - before_money + pass_reward
if buy_sell_count >= 5 and action == 0:
print("buy_sell" + str(buy_sell_count) + "回 action==" +
str(action))
reward -= (float(abs(buy_sell_count)**2))
print(reward)
elif buy_sell_count <= -5 and action == 1:
print("buy_sell" + str(buy_sell_count) + "回 action==" +
str(action))
reward -= (float(abs(buy_sell_count)**2))
print(reward)
else:
# reward 1.0がちょうど良い！
reward += 1.1
before_money = total_money
if idx % 2000 == 1000:
print("last action:" + str(action))
print("TOTAL MONEY" + str(total_money))
print("100回中passは" + str(pass_count) + "回")
# print("100回中buy_sell_countは" + str(buy_sell_count) + "回")
self.pass_count = 0
trade.draw_trading_view()
agent.save('chainerRLAgent')
# obs, reward, done, infoを返す
return X_train[price_idx], reward, False, None