版本 v1.0
### 深度学习策略的交易规则
### 策略构建步骤
### 策略的实现
在画布左侧模块列表中依次拖入输入层模块、Reshape层模块、Conv2D层模块、Reshape层模块、LSTM层模块、Dropout层模块和全连接层模块(两组)，构成深度学习网络构架，
最后通过“构建(深度学习)”模块组装各层。这里需要注意：
输入层的shape参数是 窗口滚动数据集的大小 X 因子数量 , 本例为 50 行 X 5个因子
ReShape层的参数是 窗口滚动数据集的大小 X 因子数量 X 1 ，本例为 50 行 X 5个因子 X1
Conv2D层中的 kernel_size参数是滑动窗口的尺寸，本例中使用 3行 X 5列 的窗口, 每次滑动的步长为 1行 X 1列 ， 卷积核数目为32，这里的窗口设置决定了后面ReShape层的参数
ReShape层中的target_shape 参数，这是由 窗口滚动数据集 X 因子数量 和 Conv2D层中设置的窗口尺寸以及步长决定的。本例中 50行 X 5因子 的输入数据，使用 3行 X5列 的窗口滑动取数据，
每次移动1行，共计可以得到48次数据(即可以通过滑动3行 X 5列的窗口48次来获取完整的数据)，因此target_shape= 48 X 卷积核数32
LSTM层的输出空间维度设置为卷积核数32，并设置激活函数
Dropout层是防止过度拟合采用的主动裁剪数据技术，这里设置rate 为0.8
全连接层共两层，第一层的输出空间维度与LSTM的输出维度保持一致为32，第二层将第一层的32维数据转变为1维数据输出，即获取预测的label值，此例为0到1之间的连续值，可以认为是上涨的概率。
如果当日预测的上涨概率大于0.5，则保持持仓或买入
如果当日预测的上涨概率小于0.5，则卖出股票或保持空仓。
通过 trade 模块中的初始化函数定义交易手续费和滑点，通过 context.prediction 获取每日的上涨概率预测结果；
通过 trade 模块中的主函数(handle函数)查看每日的买卖交易信号，按照买卖原则执行相应的买入/卖出操作。
可视化策略实现如下：
# 本代码由可视化策略环境自动生成 2023年9月27日 15:28
# 本代码单元只能在可视化模式下编辑。您也可以拷贝代码，粘贴到新建的代码单元或者策略，然后修改。
# 显式导入 BigQuant 相关 SDK 模块
from bigdatasource.api import DataSource
from bigdata.api.datareader import D
from biglearning.api import M
from biglearning.api import tools as T
from biglearning.module2.common.data import Outputs
import pandas as pd
import numpy as np
import math
import warnings
import datetime
from zipline.finance.commission import PerOrder
from zipline.api import get_open_orders
from zipline.api import symbol
from bigtrader.sdk import *
from bigtrader.utils.my_collections import NumPyDeque
from bigtrader.constant import OrderType
from bigtrader.constant import Direction
# 用户的自定义层需要写到字典中，比如
# {
# "MyLayer": MyLayer
# }
m6_custom_objects_bigquant_run = {
}
# Python 代码入口函数，input_1/2/3 对应三个输入端，data_1/2/3 对应三个输出端
def m2_run_bigquant_run(input_1, input_2, input_3):
test_data = input_2.read_pickle()
pred_label = input_1.read_pickle()
pred_result = pred_label.reshape(pred_label.shape[0])
dt = input_3.read_df()['date'][-1*len(pred_result):]
pred_df = pd.Series(pred_result, index=dt)
ds = DataSource.write_df(pred_df)
return Outputs(data_1=ds)
# 后处理函数，可选。输入是主函数的输出，可以在这里对数据做处理，或者返回更友好的outputs数据格式。此函数输出不会被缓存。
def m2_post_run_bigquant_run(outputs):
return outputs
# 回测引擎：初始化函数，只执行一次
def m1_initialize_bigquant_run(context):
# 加载预测数据
context.prediction = context.options['data'].read_df()
# 系统已经设置了默认的交易手续费和滑点，要修改手续费可使用如下函数
context.set_commission(PerOrder(buy_cost=0.0003, sell_cost=0.0013, min_cost=5))
# 回测引擎：每日数据处理函数，每天执行一次
def m1_handle_data_bigquant_run(context, data):
# 按日期过滤得到今日的预测数据
try:
prediction = context.prediction[data.current_dt.strftime('%Y-%m-%d')]
except KeyError as e:
return
instrument = context.instruments[0]
sid = context.symbol(instrument)
cur_position = context.portfolio.positions[sid].amount
# 交易逻辑
if prediction > 0.7 and cur_position == 0:
context.order_target_percent(context.symbol(instrument), 1)
print(data.current_dt, '买入！')
elif prediction < 0.7 and cur_position > 0:
context.order_target_percent(context.symbol(instrument), 0)
print(data.current_dt, '卖出！')
# 回测引擎：准备数据，只执行一次
def m1_prepare_bigquant_run(context):
pass
# 回测引擎：每个单位时间开始前调用一次，即每日开盘前调用一次。
def m1_before_trading_start_bigquant_run(context, data):
pass
m3 = M.dl_layer_input.v1(
shape='50,15',
batch_shape='',
dtype='float32',
sparse=False,
name=''
)
m13 = M.dl_layer_reshape.v1(
inputs=m3.data,
target_shape='50,15,1',
name=''
)
m14 = M.dl_layer_conv2d.v1(
inputs=m13.data,
filters=32,
kernel_size='3,15',
strides='1,1',
padding='valid',
data_format='channels_last',
dilation_rate='1,1',
activation='relu',
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='Zeros',
kernel_regularizer='None',
kernel_regularizer_l1=0,
kernel_regularizer_l2=0,
bias_regularizer='None',
bias_regularizer_l1=0,
bias_regularizer_l2=0,
activity_regularizer='None',
activity_regularizer_l1=0,
activity_regularizer_l2=0,
kernel_constraint='None',
bias_constraint='None',
name=''
)
m15 = M.dl_layer_reshape.v1(
inputs=m14.data,
target_shape='48,32',
name=''
)
m4 = M.dl_layer_lstm.v1(
inputs=m15.data,
units=32,
activation='tanh',
recurrent_activation='hard_sigmoid',
use_bias=True,
kernel_initializer='glorot_uniform',
recurrent_initializer='Orthogonal',
bias_initializer='Ones',
unit_forget_bias=True,
kernel_regularizer='None',
kernel_regularizer_l1=0,
kernel_regularizer_l2=0,
recurrent_regularizer='None',
recurrent_regularizer_l1=0,
recurrent_regularizer_l2=0,
bias_regularizer='None',
bias_regularizer_l1=0,
bias_regularizer_l2=0,
activity_regularizer='None',
activity_regularizer_l1=0,
activity_regularizer_l2=0,
kernel_constraint='None',
recurrent_constraint='None',
bias_constraint='None',
dropout=0,
recurrent_dropout=0,
return_sequences=False,
implementation='2',
name=''
)
m11 = M.dl_layer_dropout.v1(
inputs=m4.data,
rate=0.4,
noise_shape='',
name=''
)
m10 = M.dl_layer_dense.v1(
inputs=m11.data,
units=32,
activation='tanh',
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='Zeros',
kernel_regularizer='None',
kernel_regularizer_l1=0,
kernel_regularizer_l2=0,
bias_regularizer='None',
bias_regularizer_l1=0,
bias_regularizer_l2=0,
activity_regularizer='None',
activity_regularizer_l1=0,
activity_regularizer_l2=0,
kernel_constraint='None',
bias_constraint='None',
name=''
)
m12 = M.dl_layer_dropout.v1(
inputs=m10.data,
rate=0.8,
noise_shape='',
name=''
)
m9 = M.dl_layer_dense.v1(
inputs=m12.data,
units=1,
activation='sigmoid',
use_bias=True,
kernel_initializer='glorot_uniform',
bias_initializer='Zeros',
kernel_regularizer='None',
kernel_regularizer_l1=0,
kernel_regularizer_l2=0,
bias_regularizer='None',
bias_regularizer_l1=0,
bias_regularizer_l2=0,
activity_regularizer='None',
activity_regularizer_l1=0,
activity_regularizer_l2=0,
kernel_constraint='None',
bias_constraint='None',
name=''
)
m5 = M.dl_model_init.v1(
inputs=m3.data,
outputs=m9.data
)
m8 = M.input_features.v1(
features="""# (close_0/close_1-1)*10
# (high_0/high_1-1)*10
# (low_0/low_1-1)*10
# (open_0/open_1-1)*10
# (volume_0/volume_1-1)*10
# #号开始的表示注释，注释需单独一行
# 多个特征，每行一个，可以包含基础特征和衍生特征，特征须为本平台特征
rank_avg_amount_3
#过去 * 个交易日的平均交易额，百分比排名
rank_avg_amount_6
# 过去 * 个交易日的平均交易额，百分比排名
rank_return_3
#过去 * 个交易日的收益排名
rank_return_6
#过去 * 个交易日的收益排名
rank_return_9
#过去 * 个交易日的收益排名
return_3
#过去*个交易日的收益
avg_mf_net_amount_6
#过去6个交易日的平均净主动买入
mf_net_amount_l_0
#大单净流入
mf_net_amount_xl_0
#超大单净流入净额
mf_net_pct_main_0
#主力净流入占比
mf_net_pct_xl_0
#超大单净流入占比
rank_avg_mf_net_amount_3
#过去 * 个交易日平均净主动买入额排名
rank_avg_mf_net_amount_6
#过去 * 个交易日平均净主动买入额排名
pe_ttm_0
#市盈率
rank_pe_lyr_0
#市盈率，升序百分比排名"""
)
m24 = M.instruments.v2(
start_date='2020-01-01',
end_date='2021-12-31',
market='CN_STOCK_A',
instrument_list='600009.SHA',
max_count=0
)
m21 = M.advanced_auto_labeler.v2(
instruments=m24.data,
label_expr="""# #号开始的表示注释
# 0. 每行一个，顺序执行，从第二个开始，可以使用label字段
# 1. 可用数据字段见 https://bigquant.com/docs/develop/datasource/deprecated/history_data.html
# 添加benchmark_前缀，可使用对应的benchmark数据
# 2. 可用操作符和函数见 `表达式引擎 <https://bigquant.com/docs/develop/bigexpr/usage.html>`_
# 计算收益：5日收盘价(作为卖出价格)除以明日开盘价(作为买入价格)
where(shift(close, -5) / close -1>0,1,0)
# 过滤掉一字涨停的情况 (设置label为NaN，在后续处理和训练中会忽略NaN的label)
where(shift(high, -1) == shift(low, -1), NaN, label)
""",
start_date='',
end_date='',
benchmark='000300.SHA',
drop_na_label=True,
cast_label_int=True,
user_functions={}
)
m22 = M.general_feature_extractor.v7(
instruments=m24.data,
features=m8.data,
start_date='',
end_date='',
before_start_days=90
)
m23 = M.derived_feature_extractor.v3(
input_data=m22.data,
features=m8.data,
date_col='date',
instrument_col='instrument',
drop_na=False,
remove_extra_columns=False,
user_functions={}
)
m17 = M.join.v3(
data1=m21.data,
data2=m23.data,
on='date',
how='inner',
sort=True
)
m18 = M.dropnan.v1(
input_data=m17.data
)
m25 = M.dl_convert_to_bin.v2(
input_data=m18.data,
features=m8.data,
window_size=50,
feature_clip=15,
flatten=False,
window_along_col=''
)
m6 = M.dl_model_train.v1(
input_model=m5.data,
training_data=m25.data,
optimizer='Adam',
loss='binary_crossentropy',
metrics='accuracy',
batch_size=2048,
epochs=10,
custom_objects=m6_custom_objects_bigquant_run,
n_gpus=1,
verbose='1:输出进度条记录'
)
m28 = M.instruments.v2(
start_date=T.live_run_param('trading_date', '2022-01-01'),
end_date=T.live_run_param('trading_date', '2023-05-01'),
market='CN_STOCK_A',
instrument_list='600009.SHA',
max_count=0
)
m16 = M.general_feature_extractor.v7(
instruments=m28.data,
features=m8.data,
start_date='',
end_date='',
before_start_days=90
)
m26 = M.derived_feature_extractor.v3(
input_data=m16.data,
features=m8.data,
date_col='date',
instrument_col='instrument',
drop_na=False,
remove_extra_columns=False,
user_functions={}
)
m20 = M.dropnan.v1(
input_data=m26.data
)
m27 = M.dl_convert_to_bin.v2(
input_data=m20.data,
features=m8.data,
window_size=50,
feature_clip=15,
flatten=False,
window_along_col=''
)
m7 = M.dl_model_predict.v1(
trained_model=m6.data,
input_data=m27.data,
batch_size=10240,
n_gpus=0,
verbose='2:每个epoch输出一行记录'
)
m2 = M.cached.v3(
input_1=m7.data,
input_2=m27.data,
input_3=m20.data,
run=m2_run_bigquant_run,
post_run=m2_post_run_bigquant_run,
input_ports='',
params='{}',
output_ports=''
)
m1 = M.trade.v4(
instruments=m28.data,
options_data=m2.data_1,
start_date='',
end_date='',
initialize=m1_initialize_bigquant_run,
handle_data=m1_handle_data_bigquant_run,
prepare=m1_prepare_bigquant_run,
before_trading_start=m1_before_trading_start_bigquant_run,
volume_limit=0.025,
order_price_field_buy='twap_4',
order_price_field_sell='twap_5',
capital_base=1000000,
auto_cancel_non_tradable_orders=True,
data_frequency='daily',
price_type='真实价格',
product_type='股票',
plot_charts=True,
backtest_only=False,
benchmark='000300.HIX'
)
[2023-09-27 15:28:18.697683] INFO 自动标注(股票): 加载历史数据: 476 行 [2023-09-27 15:28:18.701885] INFO 自动标注(股票): 开始标注 ..
/var/app/enabled/bigexpr/impl/functions.py:32: FutureWarning: The `squeeze` parameter is deprecated and will be removed in a future version.
[2023-09-27 15:28:19.321113] INFO join: /y_2019, 行数=0/61, 耗时=0.070163s [2023-09-27 15:28:19.403081] INFO join: /y_2020, 行数=243/243, 耗时=0.077951s [2023-09-27 15:28:19.501073] INFO join: /y_2021, 行数=231/233, 耗时=0.089455s [2023-09-27 15:28:19.570268] INFO join: 最终行数: 474 [2023-09-27 15:28:19.730714] INFO dropnan: /y_2019, 0/0 [2023-09-27 15:28:19.822831] INFO dropnan: /y_2020, 236/243 [2023-09-27 15:28:19.909833] INFO dropnan: /y_2021, 231/231 [2023-09-27 15:28:19.974609] INFO dropnan: 行数: 467/474
2023-09-27 15:28:20.773090: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
[2023-09-27 15:28:21.180870] INFO dl_model_train: 准备训练，训练样本个数：467，迭代次数：10 Epoch 1/10 1/1 [==============================] - ETA: 0s - loss: 1.0363 - accuracy: 0.46 - 5s 5s/step - loss: 1.0363 - accuracy: 0.4604 Epoch 2/10 1/1 [==============================] - ETA: 0s - loss: 0.9353 - accuracy: 0.52 - 0s 228ms/step - loss: 0.9353 - accuracy: 0.5246 Epoch 3/10 1/1 [==============================] - ETA: 0s - loss: 1.0054 - accuracy: 0.47 - 0s 252ms/step - loss: 1.0054 - accuracy: 0.4732 Epoch 4/10 1/1 [==============================] - ETA: 0s - loss: 0.9845 - accuracy: 0.45 - 0s 245ms/step - loss: 0.9845 - accuracy: 0.4582 Epoch 5/10 1/1 [==============================] - ETA: 0s - loss: 0.9466 - accuracy: 0.46 - 0s 184ms/step - loss: 0.9466 - accuracy: 0.4625 Epoch 6/10 1/1 [==============================] - ETA: 0s - loss: 0.9311 - accuracy: 0.50 - 0s 242ms/step - loss: 0.9311 - accuracy: 0.5032 Epoch 7/10 1/1 [==============================] - ETA: 0s - loss: 0.9457 - accuracy: 0.48 - 0s 259ms/step - loss: 0.9457 - accuracy: 0.4839 Epoch 8/10 1/1 [==============================] - ETA: 0s - loss: 0.9196 - accuracy: 0.51 - 0s 221ms/step - loss: 0.9196 - accuracy: 0.5139 Epoch 9/10 1/1 [==============================] - ETA: 0s - loss: 0.9083 - accuracy: 0.51 - 0s 197ms/step - loss: 0.9083 - accuracy: 0.5182 Epoch 10/10 1/1 [==============================] - ETA: 0s - loss: 0.8850 - accuracy: 0.50 - 0s 190ms/step - loss: 0.8850 - accuracy: 0.5054 [2023-09-27 15:28:28.963580] INFO dl_model_train: 训练结束，耗时：7.78s
2023-09-27 15:28:29.837869: I tensorflow/compiler/jit/xla_gpu_device.cc:99] Not creating XLA devices, tf_xla_enable_xla_devices not set
1/1 - 0s DataSource(e4d369c2fbc84bc0a057c7f25a382849T) [2023-09-27 15:28:31.004389] INFO backtest: biglearning backtest:V8.6.3 [2023-09-27 15:28:31.009936] INFO backtest: product_type:stock by specified [2023-09-27 15:28:35.900842] INFO backtest: algo history_data=DataSource(c0a48f7bac364ffd811e3609622cc67cT) [2023-09-27 15:28:35.906282] INFO algo: TradingAlgorithm V1.8.9 [2023-09-27 15:28:36.568205] INFO algo: trading transform...
/usr/local/python3/lib/python3.8/site-packages/empyrical/stats.py:710: RuntimeWarning: divide by zero encountered in true_divide np.divide(
[2023-09-27 15:28:37.751218] INFO Performance: Simulated 320 trading days out of 320. [2023-09-27 15:28:37.756907] INFO Performance: first open: 2022-01-04 09:30:00+00:00 [2023-09-27 15:28:37.763269] INFO Performance: last close: 2023-04-28 15:00:00+00:00
/usr/local/python3/lib/python3.8/site-packages/pandas/core/generic.py:2605: PerformanceWarning:
your performance may suffer as PyTables will pickle object types that it cannot
map directly to c-types [inferred_type->mixed,key->block5_values] [items->Index(['positions', 'transactions', 'orders', 'LOG', 'TRA_FAC', 'POS_FAC',
'period_label'],
dtype='object')]
pytables.to_hdf(
/usr/local/python3/lib/python3.8/site-packages/pandas/core/generic.py:2605: PerformanceWarning:
your performance may suffer as PyTables will pickle object types that it cannot
map directly to c-types [inferred_type->mixed,key->block2_values] [items->Index(['instrument', 'suspended', 'name'], dtype='object')]
pytables.to_hdf(
/usr/local/python3/lib/python3.8/site-packages/pandas/core/indexing.py:1637: SettingWithCopyWarning:
A value is trying to be set on a copy of a slice from a DataFrame
See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
self._setitem_single_block(indexer, value, name)
[2023-09-27 15:28:39.921178] INFO: bigcharts.impl.render:render.py:408:render_chart Data is None, skip loading it to chart.