def main(args):
mode = args.mode
# mode = "test"
codes = ["600036"]
# codes = ["600036", "601998"]
# codes = args.codes
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market
# train_steps = args.train_steps
train_steps = 30000
# training_data_ratio = 0.98
training_data_ratio = args.training_data_ratio
env = Market(codes, start_date="2008-01-01", end_date="2018-01-01", **{
"market": market,
"use_sequence": True,
"scaler": MinMaxScaler,
"mix_index_state": True,
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(tf.Session(config=config), env, env.seq_length, env.data_dim, env.code_count, **{
"mode": mode,
"hidden_size": 5,
"enable_saver": True,
"train_steps": train_steps,
"enable_summary_writer": True,
"save_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot()
def main(args):
env = Market(args.codes, **{"use_sequence": True})
algorithm = Algorithm(
tf.Session(config=config),
env,
env.seq_length,
env.data_dim,
env.code_count,
**{
"mode":
args.mode,
# "mode": "test",
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", "DualAttnRNN", "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", "DualAttnRNN", "summary"),
"hidden_size":
5,
"enable_saver":
True,
"enable_summary_writer":
True
})
algorithm.run()
algorithm.eval_and_plot()
def main(args):
env = Market(args.codes)
algorithm = Algorithm(
tf.Session(config=config),
env,
env.trader.action_space,
env.data_dim,
**{
"mode":
args.mode,
# "mode": "test",
"episodes":
args.episode,
"save_path":
os.path.join(CHECKPOINTS_DIR, "RL", "DuelingDQN", "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "RL", "DuelingDQN", "summary"),
"enable_saver":
True,
"enable_summary_writer":
True
})
algorithm.run()
algorithm.eval()
algorithm.plot()
def main(args):
mode = args.mode
# mode = 'test'
codes = ["SH_index_all"]
# codes = ["600036", "601998"]
# codes = args.codes
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market
# market = 'future'
# train_steps = args.train_steps
train_steps = 30000
# training_data_ratio = 0.98
training_data_ratio = args.training_data_ratio
env = Market(codes,
start_date="2008-01-02",
end_date="2019-03-18",
**{
"market": market,
"use_sequence": True,
"seq_length": 3,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(
tf.Session(config=config), env, env.seq_length, env.data_dim,
env.code_count, **{
"mode":
mode,
"hidden_size":
64,
"layer_size":
64,
"enable_saver":
True,
"keep_prob":
0.8,
"train_steps":
train_steps,
"enable_summary_writer":
True,
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot_backtest(code=codes[0], model_name=model_name)
def main(args):
mode = args.mode
# mode = 'test'
# codes = args.codes
codes = ["600036"]
# codes = ["AU88", "RB88", "CU88", "AL88"]
# codes = ["T9999"]
market = args.market
# market = 'future'
# episode = args.episode
episode = 200
# training_data_ratio = 0.5
training_data_ratio = args.training_data_ratio
model_name = os.path.basename(__file__).split('.')[0]
env = Market(
codes,
start_date="2008-01-01",
end_date="2019-07-19",
**{
"market": market,
# "use_sequence": True,
# "mix_index_state": True,
"logger": generate_market_logger(model_name),
"training_data_ratio": training_data_ratio,
})
algorithm = Algorithm(
tf.Session(config=config), env, env.trader.action_space, env.data_dim,
**{
"mode":
mode,
"episodes":
episode,
"enable_saver":
True,
"learning_rate":
0.003,
"enable_summary_writer":
True,
"logger":
generate_algorithm_logger(model_name),
"save_path":
os.path.join(CHECKPOINTS_DIR, "RL", model_name, market, "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "RL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval()
algorithm.plot()
def market_setup_sl(args, strategy):
codes = ["SH_index"]
market = args.market
return Market(codes, start_date="2006-10-09", end_date="2019-02-27", **{
"pre_process_strategy": strategy,
"market": market,
"use_sequence": True,
"seq_length": 10,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": 0.8,
})
def main(args):
mode = args.mode
# mode = "test"
codes = ["SH_index"]
market = args.market
# train_steps = args.train_steps
# train_steps = 5000
train_steps = 1000
# training_data_ratio = 0.98
training_data_ratio = args.training_data_ratio
env = Market(codes,
start_date="2006-10-09",
end_date="2019-03-18",
**{
"market": market,
"use_sequence": True,
"seq_length": 20,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0]
print(os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"))
print(os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"))
algorithm = Algorithm(
tf.Session(config=config), env, env.seq_length, env.data_dim,
env.code_count, **{
"mode":
mode,
"hidden_size":
12,
"layer_size":
2,
"enable_saver":
False,
"train_steps":
train_steps,
"enable_summary_writer":
False,
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot_backtest(code=codes[0], model_name=model_name)
def test_nasdaq(self):
mode = self.args.mode
# mode = "test"
codes = ["nasdaq"]
market = self.args.market
# train_steps = args.train_steps
# train_steps = 5000
train_steps = 30000
# training_data_ratio = 0.98
training_data_ratio = self.args.training_data_ratio
env = Market(codes,
start_date="2008-01-02",
end_date="2019-02-01",
**{
"market": market,
"use_sequence": True,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(
tf.Session(config=config), env, env.seq_length, env.data_dim,
env.code_count, **{
"mode":
mode,
"hidden_size":
48,
"learning_rate":
0.0001,
"enable_saver":
True,
"train_steps":
train_steps,
"enable_summary_writer":
True,
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market,
"model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market,
"summary"),
})
algorithm.run()
algorithm.eval_and_plot_nasdaq_backtest()
def main(args):
mode = args.mode
#mode = 'test'
codes = args.codes
#codes = ["600036"]
#codes = ["eos_usdt"]
# codes = ["600036", "601998"]
# codes = ["AU88", "RB88", "CU88", "AL88"]
# codes = ["T9999"]
market = 'k15m'
#market = args.market
# market = 'future'
# episode = args.episode
episode = 1000
training_data_ratio = 0.95
# training_data_ratio = args.training_data_ratio
#pdb.set_trace()
model_name = os.path.basename(__file__).split('.')[0]
#env = Market(codes, start_date="2018-06-04", end_date="2018-06-12", **{
env = Market(codes, start_date=args.start, end_date=args.end, **{
"market": market,
"mix_index_state": False,
"logger": generate_market_logger(model_name),
"training_data_ratio": training_data_ratio,
})
algorithm = Algorithm(tf.Session(config=config), env, env.trader.action_space, env.data_dim, **{
"mode": mode,
"episodes": episode,
"enable_saver": True,
"learning_rate": 0.003,
"enable_summary_writer": True,
"logger": generate_algorithm_logger(model_name),
"save_path": os.path.join(CHECKPOINTS_DIR, "RL", model_name, market, "model"),
"summary_path": os.path.join(CHECKPOINTS_DIR, "RL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval()
algorithm.plot()
def main(args):
# mode = args.mode
mode = 'train'
# codes = ["600036"]
# codes = ["601398", "000651", "601998", "000002"]
codes = ["600036", "601328", "601998", "601398"]
# codes = args.codes
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market # 调用args_parser.py中的model_launcher_parser对象中的参数“stock”，指定为股票市场
# market = 'future'
# train_steps = args.train_steps
train_steps = 20000 # 训练次数
training_data_ratio = 0.8
# training_data_ratio = args.training_data_ratio
# 初始化股票市场
env = Market(codes, start_date="2008-01-01", end_date="2019-07-19", # 可选参数可接受如下字典参数
**{
"market": market,
"use_sequence": True,
"scaler": MinMaxScaler,
"mix_index_state": False,
"training_data_ratio": training_data_ratio,}
)
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(tf.Session(config=config), env, env.seq_length, env.data_dim, env.code_count, **{
"mode": mode,
"hidden_size": 5,
"enable_saver": True,
"train_steps": train_steps,
"enable_summary_writer": True,
"save_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot()
def main(args):
# mode = args.mode
mode = "train"
# codes = args.codes
# codes = ["601398"]
codes = ["600036", "601328", "601998", "601398"]
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market # default="stock"
train_steps = args.train_steps # default=100000
training_data_ratio = 0.8
# training_data_ratio = args.training_data_ratio
# env为股票市场market， market市场实例化，**可选参数传入
env = Market(codes, start_date="2008-01-01", end_date="2019-07-19", **{
"market": market,## default="stock"
"use_sequence": True,
"scaler": MinMaxScaler, # sklearn提供的缩放器
"mix_index_state": False,# 表明要混合上证指数,以结合市场趋势做更宏观的预测
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0] # 返回“TreNet，即文件名
# 算法初始化，这里是TreNet实例化， 传入一系列**可选参数
algorithm = Algorithm(tf.Session(config=config), env, env.seq_length, env.data_dim, env.code_count, **{
"mode": mode,# test
"hidden_size": 5, # 应该是5层hidden layer
"enable_saver": True,
"train_steps": train_steps,# default=100000
"enable_summary_writer": True,
"save_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot()
def init(context):
context.s1 = '600036.XSHG'
update_universe(context.s1)
context.has_save_data = False
mode = 'run'
market = 'stock'
training_data_ratio = 0.9
train_steps = 30000
base = config.get('base')
codes = ['600036']
env = Market(codes, start_date=base.get('start_date'), end_date=base.get('end_date'), **{
"market": market,
"use_sequence": True,
"scaler": MinMaxScaler,
"mix_index_state": True,
"training_data_ratio": training_data_ratio
})
model_name = 'DualAttnRNN' # os.path.basename(__file__).split('.')[0]
context.bar_list_origin = []
context.bar_list = []
context.scale = MinMaxScaler()
context.algorithm = Algorithm(
tf.Session(config=alg_config), env, env.seq_length, env.data_dim, env.code_count, **{
"mode": mode,
"hidden_size": 5,
"enable_saver": True,
"train_steps": train_steps,
"enable_summary_writer": True,
"save_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path": os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
}
)
def generator(code, start_date, end_date, market="stock", mode='trade'):
training_data_ratio = 0.8
episode = 500
env = Market(
code,
start_date=start_date,
end_date=end_date,
**{
"market": market,
# "use_sequence": True,
"logger": generate_market_logger(model_name),
"training_data_ratio": training_data_ratio,
})
return Algorithm(
tf.Session(config=config), env, env.trader.action_space,
env.data_dim, **{
"mode":
mode,
"episodes":
episode,
"enable_saver":
True,
"learning_rate":
0.003,
"enable_summary_writer":
True,
"logger":
generate_algorithm_logger(model_name),
"save_path":
os.path.join(CHECKPOINTS_DIR, "RL", model_name, market,
"model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "RL", model_name, market,
"summary"),
})
def main(args):
mode = args.mode
# mode = "test"
codes = ["SH_index_all"]
# codes = ["600036", "601998"]
# codes = args.codes
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market
# train_steps = args.train_steps
# train_steps = 5000
train_steps = 30000
# training_data_ratio = 0.98
training_data_ratio = args.training_data_ratio
env = Market(codes,
start_date="2001-01-03",
end_date="2019-03-08",
**{
"market": market,
"use_sequence": True,
"seq_length": 5,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": 0.8,
})
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(
tf.Session(config=config),
env,
env.seq_length,
env.data_dim,
env.code_count,
**{
"mode":
mode,
"layer_size":
2,
"hidden_size":
48,
# "keep_prob": 0.98, # drop out size = 1 - keep_prob
"enable_saver":
True,
"train_steps":
train_steps,
"enable_summary_writer":
True,
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
# algorithm.eval_and_plot_backtest_single(code=codes[0],
# model_name=model_name,
# MyStrategy=MyStrategy,
# DataFeed=PandasDeepLearning,
# plot=True)
algorithm.eval_and_plot_backtest(code=codes[0], model_name=model_name)
def main(args):
mode = args.mode
# mode = "test"
codes = ["SH_index_all"]
# codes = ["600036", "601998"]
# codes = args.codes
# codes = ["AU88", "RB88", "CU88", "AL88"]
market = args.market
# train_steps = args.train_steps
# train_steps = 5000
train_steps = 10000
# training_data_ratio = 0.98
training_data_ratio = args.training_data_ratio
col_order = [
'open', 'high', 'low', 'close', 'volume', 'stoch_14_3_0_3_0_0',
'stoch_14_3_0_3_0_1', 'trend_backward|close|10_5_10', 'rsi|close|14',
'roc|close|20', 'tp_score', 'on', '6m', 'nasvolume', 'nasclose',
'trend_backward|nasclose|10_5_10', 'trend|close|10_5_10'
]
'''
Index(['open', 'high', 'low', 'close', 'volume', 'on', '6m', 'nasclose',
'nasvolume', 'tp_score', 'stoch_14_3_0_3_0_0', 'stoch_14_3_0_3_0_1',
'trend_backward|close|10_5_3', 'rsi|close|14', 'roc|close|20',
'trend_backward|on|10_5_2', 'trend_backward|6m|10_5_2',
'trend_backward|nasclose|10_5_2', 'trend|close|10_5_20'],
dtype='object')
'''
env = Market(codes,
start_date="2008-01-01",
end_date="2019-03-08",
col_order=col_order,
**{
"market": market,
"use_sequence": True,
"seq_length": 5,
"scaler": MinMaxScaler(feature_range=(0, 1)),
"mix_index_state": True,
"training_data_ratio": training_data_ratio,
})
model_name = os.path.basename(__file__).split('.')[0]
algorithm = Algorithm(
tf.Session(config=config),
env,
env.seq_length,
env.data_dim,
[5, 6, 2, 3],
env.code_count,
**{
"mode":
mode,
"layer_size":
1,
"hidden_size":
16,
"keep_prob":
1, # drop out size = 1 - keep_prob
"enable_saver":
True,
"train_steps":
train_steps,
"enable_summary_writer":
True,
"save_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "model"),
"summary_path":
os.path.join(CHECKPOINTS_DIR, "SL", model_name, market, "summary"),
})
algorithm.run()
algorithm.eval_and_plot_backtest(code=codes[0], model_name=model_name)