conda create -n trademaster python=3.9

conda activate trademaster

pip install trademaster==1.2.3 torch==2.0.1 cu118 -f https://download.pytorch.org/whl/torch_stable.html

from trademaster.datasets import DJ30

dataset = DJ30(freq=’daily’, preprocessed=True)

print(dataset.feature_names) # [‘open’,’high’,’low’,’close’,’volume’,’macd’,’rsi’…]

import tushare as ts

pro = ts.pro_api(‘your_token’)

df = pro.daily(ts_code=’600519.SH’, start=’20180101′)

from trademaster.preprocessor import convert_to_ohlcv

dataset = convert_to_ohlcv(df)

import torch

print(torch.cuda.is_available()) # 输出True则成功

from torch.cuda.amp import autocast

with autocast():

agent.train()

from trademaster.strategies import BaseStrategy

class DualMAStrategy(BaseStrategy):

def __init__(self, short_window=5, long_window=20):

self.short_ma = short_window

self.long_ma = long_window

def compute_signal(self, state):

closes = state[:, :, 3]

ma5 = closes[-self.short_ma:].mean(axis=0)

ma20 = closes[-self.long_ma:].mean(axis=0)

return (ma5 > ma20).astype(float)

from trademaster.agents import PPOAgent

config = {

“n_steps”: 2048,

“batch_size”: 64,

“gae_lambda”: 0.95,

“clip_range”: 0.2,

}

def custom_reward(portfolio_return, volatility):

return portfolio_return – 0.3 * volatility

agent = PPOAgent(env, reward_fn=custom_reward, **config)

from trademaster.pretrained import load_eiie

model = load_eiie(dataset=’SP500′, version=’v2.1′)

model.freeze_layers(exclude=[‘lstm’])

model.fit(train_dataset, epochs=50, lr=1e-4)

report = env.get_performance_report()

print(f”””

■ 收益指标:

累计收益率: {report[‘total_return’]:.2%}

年化收益率: {report[‘annual_return’]:.2%}

■ 风险指标:

最大回撤: {report[‘max_drawdown’]:.2%}

波动率: {report[‘volatility’]:.2%}

“””)

from trademaster.analysis import plot_cumulative_returns

plot_cumulative_returns({

‘双均线’: baseline_env,

‘PPO’: ppo_env,

‘EIIE’: eiie_env

})

from trademaster.brokers import QMTBroker

qmt = QMTBroker(

path=’C:/国金QMT交易端’,

account_id=’123456′,

latency=0.001

)

orders = [

{‘symbol’: ‘600519.SH’, ‘quantity’: 1000, ‘side’: ‘buy’},

{‘symbol’: ‘000001.SZ’, ‘quantity’: 2000, ‘side’: ‘sell’}

]

qmt.batch_order(orders)

FROM nvidia/cuda:12.2.0-devel-ubuntu22.04

RUN apt-get update && apt-get install -y python3.9 pip

RUN pip install trademaster pyfolio

ENV CUDA_VISIBLE_DEVICES=0

CMD [“python”, “/app/main.py”]

apiVersion: apps/v1

kind: Deployment

spec:

template:

spec:

containers:

– name: strategy

image: trademaster:v1.0

resources:

limits:

nvidia.com/gpu: 1

def process_orderbook(msg):

return {

‘timestamp’: msg[‘timestamp’],

‘bids’: [(p, s) for p, s in zip(msg[‘bid_prices’], msg[‘bid_sizes’])],

‘asks’: [(p, s) for p, s in zip(msg[‘ask_prices’], msg[‘ask_sizes’])]

}

def pressure_metric(orderbook):

bid_vol = sum([s for p, s in orderbook[‘bids’][:5]])

ask_vol = sum([s for p, s in orderbook[‘asks’][:5]])

return (bid_vol – ask_vol) / (bid_vol + ask_vol)

from numba import jit

@jit(nopython=True)

def fast_correlation(x, y):

sum_x = sum(x)

sum_y = sum(y)

sum_xy = np.dot(x, y)

# … 快速计算相关系数

from trademaster.validations import walk_forward_validation

wf_scores = walk_forward_validation(

dataset, window_size=252, test_size=63

)

print(f”策略稳定性得分: {np.mean(wf_scores):.2f}”)

try:

broker.execute_order(order)

except MarketClosedError:

self.logger.warning(“市场已收盘，订单取消”)

except NetworkException as e:

self.retry_queue.append(order)