def save_eval_result(self):
tools.listToCsv(self.result_folder+'log.csv',self.train_ave_profit_list,self.ave_Q_list,self.ave_reward_list,self.test_ave_profit_list,self.epsilon_list)
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('ave_max_Q')
axis2.set_ylabel('epsilon')
axis1.plot(self.ave_Q_list, label = "ave_max_Q")
axis1.legend(loc = 'upper left')
axis2.plot(self.epsilon_list, label = 'epsilon', color = 'g')
axis2.legend()
filename = self.result_folder + "log_ave_max_Q.png"
plt.grid(which='major')
plt.savefig(filename)
plt.close()
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('ave_reward')
axis2.set_ylabel('epsilon')
axis1.plot(self.ave_reward_list, label = "ave_reward")
reward_ema = ta.EMA(np.array(self.ave_reward_list,dtype='f8'),timeperiod=10)
axis1.plot(reward_ema, color = 'r')
axis1.legend(loc = 'upper left')
axis2.plot(self.epsilon_list, label = 'epsilon', color = 'g')
axis2.legend()
filename = self.result_folder + "log_ave_reward.png"
plt.grid(which='major')
plt.savefig(filename)
plt.close()
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('ave_train_profit')
axis2.set_ylabel('epsilon')
axis1.plot(self.train_ave_profit_list, label = "ave_train_profit")
train_ema = ta.EMA(np.array(self.train_ave_profit_list,dtype='f8'),timeperiod=10)
axis1.plot(train_ema, color = 'r')
axis1.legend(loc = 'upper left')
axis2.plot(self.epsilon_list, label = 'epsilon', color = 'g')
axis2.legend()
filename = self.result_folder + "log_ave_train_profit.png"
plt.grid(which='major')
plt.savefig(filename)
plt.close()
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('ave_test_profit')
axis2.set_ylabel('epsilon')
axis1.plot(self.test_ave_profit_list, label = "ave_test_profit")
test_ema = ta.EMA(np.array(self.test_ave_profit_list,dtype='f8'),timeperiod=10)
axis1.plot(test_ema, color = 'r')
axis1.legend(loc = 'upper left')
axis2.plot(self.epsilon_list, label = 'epsilon', color = 'g')
axis2.legend()
filename = self.result_folder + "log_ave_test_profit.png"
plt.savefig(filename)
plt.close()
Agent.DQN.reset_optimizer()
stock_agent = env_stockmarket.Stock_agent(Agent, args.action_split_number)
try:
testdata, testprice = market.get_testData(f, args.input_num)
#testdata, testprice = market.get_trainData(f,END_TRAIN_DAY,args.input_num)
except:
print 'skip', f
continue
profit_ratio, proper, order, stocks, price, Q_list, ave_buyprice_list, reward_list = stock_agent.trading_test(
args.input_num, testprice, testdata)
profit_list.append(profit_ratio)
tools.listToCsv(folder + str(f).replace(".CSV", "") + '.csv', price,
proper, order, stocks, ave_buyprice_list, reward_list)
buy_order, sell_order = tools.order2buysell(order, price)
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('price')
axis1.set_ylabel('buy')
axis1.set_ylabel('sell')
axis2.set_ylabel('property')
axis1.plot(price, label="price")
axis1.plot(buy_order, 'o', label='buy point')
axis1.plot(sell_order, '^', label='sell point')
axis1.legend(loc='upper left')
def log_result():
# Logger
global train_loss_list, mean_error_list
train_count = 0
train_cur_loss = 0
train_cur_accuracy = 0
begin_at = time.time()
val_begin_at = None
while True:
result = res_q.get()
if result == 'end':
print(file=sys.stderr)
break
elif result == 'train':
print(file=sys.stderr)
train = True
if val_begin_at is not None:
begin_at += time.time() - val_begin_at
val_begin_at = None
continue
elif result == 'val':
print(file=sys.stderr)
train = False
val_count = val_loss = val_accuracy = 0
val_begin_at = time.time()
continue
loss, accuracy = result
if train:
train_count += 1
duration = time.time() - begin_at
throughput = train_count * args.batchsize / duration
sys.stderr.write(
'\rtrain {} updates ({} samples) time: {} ({} images/sec)'
.format(train_count, train_count * args.batchsize,
datetime.timedelta(seconds=duration), throughput))
train_cur_loss += loss
train_cur_accuracy += accuracy
if train_count % 10 == 0:
mean_loss = train_cur_loss / 10
train_loss_list.append(mean_loss)
mean_error = 1 - train_cur_accuracy / 10
mean_error_list.append(mean_error)
tools.listToCsv(folder+'trainloss.csv',train_loss_list,mean_error_list)
print(file=sys.stderr)
print(json.dumps({'type': 'train', 'iteration': train_count,
'error': mean_error, 'loss': mean_loss}))
sys.stdout.flush()
train_cur_loss = 0
train_cur_accuracy = 0
else:
val_count += args.batchsize
duration = time.time() - val_begin_at
throughput = val_count / duration
sys.stderr.write(
'\rval {} batches ({} samples) time: {} ({} images/sec)'
.format(val_count / args.batchsize, val_count,
datetime.timedelta(seconds=duration), throughput))
val_loss += loss
val_accuracy += accuracy
if val_count == 10000:
mean_loss = val_loss * args.batchsize / 10000
mean_error = 1 - val_accuracy * args.batchsize / 10000
print(file=sys.stderr)
print(json.dumps({'type': 'val', 'iteration': train_count,
'error': mean_error, 'loss': mean_loss}))
sys.stdout.flush()
for f in files:
print f
stock_agent = env_stockmarket.Stock_agent(Agent)
try:
testdata,testprice = market.get_testData(f,args.input_num)
#testdata, testprice = market.get_trainData(f,END_TRAIN_DAY,args.input_num)
except:
print 'skip',f
continue
profit_ratio, proper, order, stocks, price = stock_agent.trading_test(args.input_num,testprice,testdata)
profit_list.append(profit_ratio)
tools.listToCsv(folder+str(f).replace(".CSV", "")+'.csv', price, proper, order,stocks)
buy_order, sell_order = tools.order2buysell(order,price)
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('price')
axis1.set_ylabel('buy')
axis1.set_ylabel('sell')
axis2.set_ylabel('property')
axis1.plot(price, label = "price")
axis1.plot(buy_order,'o',label='buy')
axis1.plot(sell_order,'^',label='sell')
axis1.legend(loc = 'upper left')
axis2.plot(proper, label = 'property', color = 'g')
inputlist = row[:-output_num-2]
output = row[-output_num-2]
inputlist = np.array([inputlist]).astype(np.float32)
outputlist.append(output)
y = model.predict(xp.asarray(inputlist))
predictlist.append(y.data[0,0])
rec.append((output - y.data[0,0])*(output - y.data[0,0]))
price = tools.getClose(args.code,START_TEST_DAY)
#print len(test)
#print len(outputlist), len(predictlist) ,len(price)
ema_list = ta.EMA(np.array(rec,dtype='f8'),timeperiod=60)
tools.listToCsv(folder+'visuallizer' + str(args.code)+'.csv',price[:-NEXT_DAY+1],outputlist,predictlist,rec,ema_list)
#可視化
#2軸使用
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('price')
axis2.set_ylabel('move ratio')
axis1.plot(price, label = "price")
axis1.legend(loc = 'upper left')
axis2.plot(outputlist, label = 'True', color = 'g')
axis2.plot(predictlist, label = 'predict', color = 'r')
axis2.plot(ema_list,label='ema',color='y')
#axis2.plot(error, label = 'square error', color = 'm')
axis2.legend()
plt.show()
print 'finished!!'
for row in test:
inputlist = row[:-output_num-2]
output = row[-output_num-2]
inputlist = np.array([inputlist]).astype(np.float32)
y = model_3.predict(xp.asarray(inputlist))
rec_3.append((output - y.data[0,0])*(output - y.data[0,0]))
price = tools.getClose(args.code,START_TEST_DAY)
#print len(test)
#print len(outputlist), len(predictlist) ,len(price)
sma_list_1 = ta.SMA(np.array(rec_1,dtype='f8'),timeperiod=60)
sma_list_2 = ta.SMA(np.array(rec_2,dtype='f8'),timeperiod=60)
sma_list_3 = ta.SMA(np.array(rec_3,dtype='f8'),timeperiod=60)
tools.listToCsv(folder+'visuallizer_multi' + str(args.code)+'.csv',price[:-NEXT_DAY+1],sma_list_1,sma_list_2,sma_list_3)
#可視化
#2軸使用
pic_name = folder + 'visuallizer_multi' + str(args.code)+'.png'
fig, axis1 = plt.subplots()
axis2 = axis1.twinx()
axis1.set_ylabel('price')
axis2.set_ylabel('move ratio')
axis1.plot(price, label = "price")
axis1.legend(loc = 'upper left')
axis2.plot(sma_list_1,label='model_1',color='y')
axis2.plot(sma_list_2,label='model_2',color='c')
axis2.plot(sma_list_3,label='model_3',color='m')
#axis2.plot(error, label = 'square error', color = 'm')
axis2.legend()
def main(files):
start_trading_day = 20000104
end_trading_day = 20081230
#start_trading_day = 20090105
#end_trading_day = 20101229
meigara_count = 0
profit_ratio_rsi_list = []
profit_ratio_macd_list = []
profit_ratio_gd_list = []
profit_ratio_stoch_list = []
profit_ratio_rsi_list = []
bh_profit_ratio_list = []
for f in tqdm(files):
#print f
_property = 0 #総資産
money = 1000000 #所持金
filepath = args.data_folder + '/' + f
#株価データの読み込み
_time, _open, _max, _min, _close, _volume, _keisu, _shihon = make_dataset.readfile(
filepath)
try:
iday = _time.index(start_trading_day)
eday = _time.index(end_trading_day)
except:
#print "can't find start_test_day"
continue #start_trading_dayが見つからなければ次のファイルへ
point_rsi = getStrategy_RSI(start_trading_day, end_trading_day, _time,
_close)
point_macd = getStrategy_MACD(start_trading_day, end_trading_day,
_time, _close)
point_gd = getStrategy_GD(start_trading_day, end_trading_day, _time,
_close)
point_stoch = getStrategy_STOCH(start_trading_day, end_trading_day,
_time, _close, _max, _min)
#売買開始日からスライス
_time = _time[iday:eday]
_close = _close[iday:eday]
#buy&holdの利益率を計算
bh_profit_ratio = float((_close[-1] - _close[0]) / _close[0]) * 100
bh_profit_ratio_list.append(bh_profit_ratio)
profit_ratio = trading(money, point_rsi, _close)[0]
profit_ratio_rsi_list.append(profit_ratio)
profit_ratio = trading(money, point_macd, _close)[0]
profit_ratio_macd_list.append(profit_ratio)
profit_ratio = trading(money, point_gd, _close)[0]
profit_ratio_gd_list.append(profit_ratio)
profit_ratio = trading(money, point_stoch, _close)[0]
profit_ratio_stoch_list.append(profit_ratio)
meigara_count += 1
#print meigara_count
print "RSI profit average is = ", sum(profit_ratio_rsi_list) / len(
profit_ratio_rsi_list)
print "MACD profit average is = ", sum(profit_ratio_macd_list) / len(
profit_ratio_macd_list)
print "GD profit average is = ", sum(profit_ratio_gd_list) / len(
profit_ratio_gd_list)
print "STOCH profit average is = ", sum(profit_ratio_stoch_list) / len(
profit_ratio_stoch_list)
print 'buy&hold profit = ', sum(bh_profit_ratio_list) / len(
bh_profit_ratio_list)
print "all meigara is %d" % meigara_count
tools.listToCsv('profit_ratio_list.csv', bh_profit_ratio_list,
profit_ratio_gd_list, profit_ratio_macd_list,
profit_ratio_rsi_list, profit_ratio_stoch_list)