克隆策略

练习题¶

修改Transformer模型的Decoder模块，并训练一个自定义的Transformer模型用于选股，并进行回测。

参考¶

Transformer Decoder可以改进的方向
- MLP Decoder
- GRU/LSTM Decoder
- CNN Decoder
Code
- https://github.com/microsoft/qlib/blob/main/qlib/contrib/model/pytorch_localformer_ts.py
Paper
- Transformer-Based Capsule Network For Stock Movements Prediction
- 1D Convolutional Neural Networks and Applications – A Survey

In [1]:

import math
import torch
import torch.nn as nn

from bigmodels.models.base import BaseModel

In [2]:

class PositionalEncoding(nn.Module):
    def __init__(self, d_model, max_len=1000):
        super(PositionalEncoding, self).__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-math.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0).transpose(0, 1)
        self.register_buffer("pe", pe)

    def forward(self, x):
        # [T, N, F]
        return x + self.pe[: x.size(0), :]

In [3]:

class Transformer(BaseModel):
    """
    Transformer:Attention is all you need
    paper: https://arxiv.org/abs/1706.03762
    """

    def __init__(
        self,
        input_dim=98,
        output_dim=1,
        max_seq=5,
        embed_dim=128,
        nhead=8,
        num_layers=4,
        dropout=0.1,
    ):
        super(Transformer, self).__init__()
        self.input_dim = input_dim
        self.output_dim = output_dim

        self.feature_layer = nn.Linear(input_dim, embed_dim)
        # self.feature_pos = nn.Embedding(max_seq, embed_dim)
        
        # update position encoder with Positional
        self.pos_encoder = PositionalEncoding(embed_dim)

        encoder_layer = nn.TransformerEncoderLayer(d_model=embed_dim, nhead=nhead, dropout=dropout)
        self.encoder = nn.TransformerEncoder(encoder_layer, num_layers=num_layers)

        # TODO, FNN for Decoder
        self.out = nn.Linear(embed_dim, output_dim)
        self._reset_parameters()

    def _reset_parameters(self):
        r"""Initiate parameters in the model."""
        for p in self.parameters():
            if p.dim() > 1:
                nn.init.xavier_uniform_(p)

    def forward(self, x):
        x = self.feature_layer(x)

        # x = [bs, window, feature] -> [window, bs, feature]
        x = x.permute(1, 0, 2)

        x = self.pos_encoder(x)

        src_mask = None
        x = self.encoder(x, src_mask)
        x = x.permute(1, 0, 2)
        
        # update Decoder
        # x = [bs, window, embed_dim]
        x = x[:, -1, :]
        x = self.out(x)
        
        return x.squeeze()

In [4]:

# 本代码由可视化策略环境自动生成 2021年10月29日 17:02
# 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。


# 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 m33_run_bigquant_run(input_1, input_2, input_3):
    # 示例代码如下。在这里编写您的代码
    from sklearn.model_selection import train_test_split
    # train data
    train_data = input_1.read()
    x_train, x_val, y_train, y_val = train_test_split(train_data["x"], train_data['y'], shuffle=True, random_state=2021)
    # val data
    test_data = input_2.read()
    x_test = test_data["x"]
    
    model = Transformer(input_dim=98, embed_dim=256, nhead=8, num_layers=6, dropout=0.1)
    model.compile(device="cuda:0")
    model.fit(x_train, y_train, val_data=(x_val, y_val), batch_size=2048, epochs=10, verbose=1, num_workers=2)
    
    # model.fit(train_data["x"], train_data['y'], batch_size=1024, epochs=2, verbose=1, num_workers=2)
    output = model.predict(x_test)
    
    data_1 = DataSource.write_pickle(output)
    return Outputs(data_1=data_1, data_2=None, data_3=None)
# 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 m33_post_run_bigquant_run(outputs):
    return outputs

# 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 m41_run_bigquant_run(input_1, input_2, input_3):
    # 示例代码如下。在这里编写您的代码
    pred_label = input_1.read_pickle()
    
    df = input_2.read_df()
    df = pd.DataFrame({'pred_label':pred_label[:], 'instrument':df.instrument, 'date':df.date})
    df.sort_values(['date','pred_label'],inplace=True, ascending=[True,False])
    return Outputs(data_1=DataSource.write_df(df), data_2=None, data_3=None)

# 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 m41_post_run_bigquant_run(outputs):
    return outputs

# 回测引擎：初始化函数，只执行一次
def m42_initialize_bigquant_run(context):
    # 加载预测数据
    context.ranker_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=0.001, sell_cost=0.001, min_cost=5))
    # 预测数据，通过options传入进来，使用 read_df 函数，加载到内存 (DataFrame)
    # 设置买入的股票数量，这里买入预测股票列表排名靠前的5只
    stock_count = 20
    # 每只的股票的权重，如下的权重分配会使得靠前的股票分配多一点的资金，[0.339160, 0.213986, 0.169580, ..]
    context.stock_weights = T.norm([1 / math.log(i + 2) for i in range(0, stock_count)])
    # 设置每只股票占用的最大资金比例
    context.max_cash_per_instrument = 0.2
    context.options['hold_days'] = 5

# 回测引擎：每日数据处理函数，每天执行一次
def m42_handle_data_bigquant_run(context, data):
    # 按日期过滤得到今日的预测数据
    ranker_prediction = context.ranker_prediction[
        context.ranker_prediction.date == data.current_dt.strftime('%Y-%m-%d')]

    # 1. 资金分配
    # 平均持仓时间是hold_days，每日都将买入股票，每日预期使用 1/hold_days 的资金
    # 实际操作中，会存在一定的买入误差，所以在前hold_days天，等量使用资金；之后，尽量使用剩余资金（这里设置最多用等量的1.5倍）
    is_staging = context.trading_day_index < context.options['hold_days'] # 是否在建仓期间（前 hold_days 天）
    cash_avg = context.portfolio.portfolio_value / context.options['hold_days']
    cash_for_buy = min(context.portfolio.cash, (1 if is_staging else 1.5) * cash_avg)
    cash_for_sell = cash_avg - (context.portfolio.cash - cash_for_buy)
    positions = {e.symbol: p.amount * p.last_sale_price
                 for e, p in context.perf_tracker.position_tracker.positions.items()}

    # 2. 生成卖出订单：hold_days天之后才开始卖出；对持仓的股票，按机器学习算法预测的排序末位淘汰
    if not is_staging and cash_for_sell > 0:
        equities = {e.symbol: e for e, p in context.perf_tracker.position_tracker.positions.items()}
        instruments = list(reversed(list(ranker_prediction.instrument[ranker_prediction.instrument.apply(
                lambda x: x in equities and not context.has_unfinished_sell_order(equities[x]))])))
        # print('rank order for sell %s' % instruments)
        for instrument in instruments:
            context.order_target(context.symbol(instrument), 0)
            cash_for_sell -= positions[instrument]
            if cash_for_sell <= 0:
                break

    # 3. 生成买入订单：按机器学习算法预测的排序，买入前面的stock_count只股票
    buy_cash_weights = context.stock_weights
    buy_instruments = list(ranker_prediction.instrument[:len(buy_cash_weights)])
    max_cash_per_instrument = context.portfolio.portfolio_value * context.max_cash_per_instrument
    for i, instrument in enumerate(buy_instruments):
        cash = cash_for_buy * buy_cash_weights[i]
        if cash > max_cash_per_instrument - positions.get(instrument, 0):
            # 确保股票持仓量不会超过每次股票最大的占用资金量
            cash = max_cash_per_instrument - positions.get(instrument, 0)
        if cash > 0:
            context.order_value(context.symbol(instrument), cash)

# 回测引擎：准备数据，只执行一次
def m42_prepare_bigquant_run(context):
    pass


m22 = M.instruments.v2(
    start_date='2010-01-01',
    end_date='2017-12-31',
    market='CN_STOCK_A',
    instrument_list='',
    max_count=0
)

m23 = M.advanced_auto_labeler.v2(
    instruments=m22.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/data_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/big_expr.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)
shift(close, -5) / shift(open, -1)-1

# 极值处理：用1%和99%分位的值做clip
clip(label, all_quantile(label, 0.01), all_quantile(label, 0.99))

# 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=False
)

m1 = M.standardlize.v8(
    input_1=m23.data,
    columns_input='label'
)

m24 = M.input_features.v1(
    features="""close_0
open_0
high_0
low_0 
amount_0
turn_0 
return_0
 
close_1
open_1
high_1
low_1
return_1
amount_1
turn_1
 
close_2
open_2
high_2
low_2
amount_2
turn_2
return_2
 
close_3
open_3
high_3
low_3
amount_3
turn_3
return_3
 
close_4
open_4
high_4
low_4
amount_4
turn_4
return_4
 
mean(close_0, 5)
mean(low_0, 5)
mean(open_0, 5)
mean(high_0, 5)
mean(turn_0, 5)
mean(amount_0, 5)
mean(return_0, 5)
 
ts_max(close_0, 5)
ts_max(low_0, 5)
ts_max(open_0, 5)
ts_max(high_0, 5)
ts_max(turn_0, 5)
ts_max(amount_0, 5)
ts_max(return_0, 5)
 
ts_min(close_0, 5)
ts_min(low_0, 5)
ts_min(open_0, 5)
ts_min(high_0, 5)
ts_min(turn_0, 5)
ts_min(amount_0, 5)
ts_min(return_0, 5) 
 
std(close_0, 5)
std(low_0, 5)
std(open_0, 5)
std(high_0, 5)
std(turn_0, 5)
std(amount_0, 5)
std(return_0, 5)
 
ts_rank(close_0, 5)
ts_rank(low_0, 5)
ts_rank(open_0, 5)
ts_rank(high_0, 5)
ts_rank(turn_0, 5)
ts_rank(amount_0, 5)
ts_rank(return_0, 5)
 
decay_linear(close_0, 5)
decay_linear(low_0, 5)
decay_linear(open_0, 5)
decay_linear(high_0, 5)
decay_linear(turn_0, 5)
decay_linear(amount_0, 5)
decay_linear(return_0, 5)
 
correlation(volume_0, return_0, 5)
correlation(volume_0, high_0, 5)
correlation(volume_0, low_0, 5)
correlation(volume_0, close_0, 5)
correlation(volume_0, open_0, 5)
correlation(volume_0, turn_0, 5)
  
correlation(return_0, high_0, 5)
correlation(return_0, low_0, 5)
correlation(return_0, close_0, 5)
correlation(return_0, open_0, 5)
correlation(return_0, turn_0, 5)
 
correlation(high_0, low_0, 5)
correlation(high_0, close_0, 5)
correlation(high_0, open_0, 5)
correlation(high_0, turn_0, 5)
 
correlation(low_0, close_0, 5)
correlation(low_0, open_0, 5)
correlation(low_0, turn_0, 5)
 
correlation(close_0, open_0, 5)
correlation(close_0, turn_0, 5)

correlation(open_0, turn_0, 5)"""
)

m27 = M.general_feature_extractor.v7(
    instruments=m22.data,
    features=m24.data,
    start_date='',
    end_date='',
    before_start_days=10
)

m28 = M.derived_feature_extractor.v3(
    input_data=m27.data,
    features=m24.data,
    date_col='date',
    instrument_col='instrument',
    drop_na=True,
    remove_extra_columns=False
)

m34 = M.standardlize.v8(
    input_1=m28.data,
    input_2=m24.data,
    columns_input='[]'
)

m35 = M.fillnan.v1(
    input_data=m34.data,
    features=m24.data,
    fill_value='0.0'
)

m25 = M.join.v3(
    data1=m1.data,
    data2=m35.data,
    on='date,instrument',
    how='inner',
    sort=False
)

m2 = M.dl_convert_to_bin.v2(
    input_data=m25.data,
    features=m24.data,
    window_size=5,
    feature_clip=3,
    flatten=False,
    window_along_col='instrument'
)

m26 = M.instruments.v2(
    start_date='2018-01-01',
    end_date='2021-07-01',
    market='CN_STOCK_A',
    instrument_list='',
    max_count=0
)

m29 = M.general_feature_extractor.v7(
    instruments=m26.data,
    features=m24.data,
    start_date='',
    end_date='',
    before_start_days=10
)

m30 = M.derived_feature_extractor.v3(
    input_data=m29.data,
    features=m24.data,
    date_col='date',
    instrument_col='instrument',
    drop_na=True,
    remove_extra_columns=False
)

m37 = M.standardlize.v8(
    input_1=m30.data,
    input_2=m24.data,
    columns_input='[]'
)

m36 = M.fillnan.v1(
    input_data=m37.data,
    features=m24.data,
    fill_value='0.0'
)

m32 = M.dl_convert_to_bin.v2(
    input_data=m36.data,
    features=m24.data,
    window_size=5,
    feature_clip=3,
    flatten=False,
    window_along_col='instrument'
)

m33 = M.cached.v3(
    input_1=m2.data,
    input_2=m32.data,
    run=m33_run_bigquant_run,
    post_run=m33_post_run_bigquant_run,
    input_ports='',
    params='{}',
    output_ports='',
    m_cached=False
)

m41 = M.cached.v3(
    input_1=m33.data_1,
    input_2=m30.data,
    run=m41_run_bigquant_run,
    post_run=m41_post_run_bigquant_run,
    input_ports='',
    params='{}',
    output_ports='',
    m_cached=False
)

m42 = M.trade.v4(
    instruments=m26.data,
    options_data=m41.data_1,
    start_date='',
    end_date='',
    initialize=m42_initialize_bigquant_run,
    handle_data=m42_handle_data_bigquant_run,
    prepare=m42_prepare_bigquant_run,
    volume_limit=0.025,
    order_price_field_buy='open',
    order_price_field_sell='close',
    capital_base=1000000,
    auto_cancel_non_tradable_orders=True,
    data_frequency='daily',
    price_type='后复权',
    product_type='股票',
    plot_charts=True,
    backtest_only=False,
    benchmark='000300.SHA'
)

[2021-10-29 16:16:14.113072] INFO: moduleinvoker: instruments.v2 开始运行..

[2021-10-29 16:16:14.138630] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.140300] INFO: moduleinvoker: instruments.v2 运行完成[0.027252s].

[2021-10-29 16:16:14.149162] INFO: moduleinvoker: advanced_auto_labeler.v2 开始运行..

[2021-10-29 16:16:14.160312] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.162105] INFO: moduleinvoker: advanced_auto_labeler.v2 运行完成[0.01294s].

[2021-10-29 16:16:14.168630] INFO: moduleinvoker: standardlize.v8 开始运行..

[2021-10-29 16:16:14.179578] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.180862] INFO: moduleinvoker: standardlize.v8 运行完成[0.01223s].

[2021-10-29 16:16:14.185089] INFO: moduleinvoker: input_features.v1 开始运行..

[2021-10-29 16:16:14.190305] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.191514] INFO: moduleinvoker: input_features.v1 运行完成[0.006426s].

[2021-10-29 16:16:14.267559] INFO: moduleinvoker: general_feature_extractor.v7 开始运行..

[2021-10-29 16:16:14.278509] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.280189] INFO: moduleinvoker: general_feature_extractor.v7 运行完成[0.012654s].

[2021-10-29 16:16:14.289585] INFO: moduleinvoker: derived_feature_extractor.v3 开始运行..

[2021-10-29 16:16:14.295948] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.297432] INFO: moduleinvoker: derived_feature_extractor.v3 运行完成[0.007845s].

[2021-10-29 16:16:14.322195] INFO: moduleinvoker: standardlize.v8 开始运行..

[2021-10-29 16:16:14.329689] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.331050] INFO: moduleinvoker: standardlize.v8 运行完成[0.008852s].

[2021-10-29 16:16:14.339325] INFO: moduleinvoker: fillnan.v1 开始运行..

[2021-10-29 16:16:14.348285] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.349580] INFO: moduleinvoker: fillnan.v1 运行完成[0.010253s].

[2021-10-29 16:16:14.358061] INFO: moduleinvoker: join.v3 开始运行..

[2021-10-29 16:16:14.401273] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.403190] INFO: moduleinvoker: join.v3 运行完成[0.045116s].

[2021-10-29 16:16:14.417725] INFO: moduleinvoker: dl_convert_to_bin.v2 开始运行..

[2021-10-29 16:16:14.426664] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.428153] INFO: moduleinvoker: dl_convert_to_bin.v2 运行完成[0.010442s].

[2021-10-29 16:16:14.432863] INFO: moduleinvoker: instruments.v2 开始运行..

[2021-10-29 16:16:14.441012] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.442354] INFO: moduleinvoker: instruments.v2 运行完成[0.00949s].

[2021-10-29 16:16:14.452873] INFO: moduleinvoker: general_feature_extractor.v7 开始运行..

[2021-10-29 16:16:14.464469] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.465932] INFO: moduleinvoker: general_feature_extractor.v7 运行完成[0.01306s].

[2021-10-29 16:16:14.472398] INFO: moduleinvoker: derived_feature_extractor.v3 开始运行..

[2021-10-29 16:16:14.495139] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.500551] INFO: moduleinvoker: derived_feature_extractor.v3 运行完成[0.028139s].

[2021-10-29 16:16:14.505484] INFO: moduleinvoker: standardlize.v8 开始运行..

[2021-10-29 16:16:14.518520] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.519816] INFO: moduleinvoker: standardlize.v8 运行完成[0.014332s].

[2021-10-29 16:16:14.526570] INFO: moduleinvoker: fillnan.v1 开始运行..

[2021-10-29 16:16:14.533701] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.534929] INFO: moduleinvoker: fillnan.v1 运行完成[0.008356s].

[2021-10-29 16:16:14.545440] INFO: moduleinvoker: dl_convert_to_bin.v2 开始运行..

[2021-10-29 16:16:14.552787] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:16:14.554001] INFO: moduleinvoker: dl_convert_to_bin.v2 运行完成[0.008564s].

[2021-10-29 16:16:14.563062] INFO: moduleinvoker: cached.v3 开始运行..

[2021-10-29 16:59:20.210136] INFO: moduleinvoker: cached.v3 运行完成[2585.647047s].

[2021-10-29 16:59:20.220305] INFO: moduleinvoker: cached.v3 开始运行..

[2021-10-29 16:59:42.505209] INFO: moduleinvoker: cached.v3 运行完成[22.284891s].

[2021-10-29 16:59:44.279469] INFO: moduleinvoker: backtest.v8 开始运行..

[2021-10-29 16:59:44.301467] INFO: backtest: biglearning backtest:V8.5.0

[2021-10-29 16:59:44.303299] INFO: backtest: product_type:stock by specified

[2021-10-29 16:59:44.429730] INFO: moduleinvoker: cached.v2 开始运行..

[2021-10-29 16:59:44.444428] INFO: moduleinvoker: 命中缓存

[2021-10-29 16:59:44.445962] INFO: moduleinvoker: cached.v2 运行完成[0.016258s].

[2021-10-29 16:59:47.999698] INFO: algo: TradingAlgorithm V1.8.5

[2021-10-29 16:59:49.664568] INFO: algo: trading transform...

[2021-10-29 17:00:51.180704] INFO: Performance: Simulated 849 trading days out of 849.

[2021-10-29 17:00:51.182458] INFO: Performance: first open: 2018-01-02 09:30:00+00:00

[2021-10-29 17:00:51.183786] INFO: Performance: last close: 2021-07-01 15:00:00+00:00

[2021-10-29 17:01:03.372516] INFO: moduleinvoker: backtest.v8 运行完成[79.09305s].

[2021-10-29 17:01:03.374305] INFO: moduleinvoker: trade.v4 运行完成[80.846919s].

epoch 0   |  train_loss 1.04397|  vall_loss 0.98657|  0:04:07s
epoch 1   |  train_loss 0.98324|  vall_loss 0.98302|  0:08:13s
epoch 2   |  train_loss 0.98024|  vall_loss 0.98060|  0:12:20s
epoch 3   |  train_loss 0.97917|  vall_loss 0.98368|  0:16:28s
epoch 4   |  train_loss 0.97855|  vall_loss 0.98163|  0:20:34s
epoch 5   |  train_loss 0.97704|  vall_loss 0.98008|  0:24:41s
epoch 6   |  train_loss 0.97423|  vall_loss 0.97625|  0:28:47s
epoch 7   |  train_loss 0.96945|  vall_loss 0.97266|  0:32:54s
epoch 8   |  train_loss 0.96386|  vall_loss 0.97054|  0:37:00s
epoch 9   |  train_loss 0.96000|  vall_loss 0.97070|  0:41:07s
best loss: 0.970537 @ 8

收益率193.44%
年化收益率37.65%
基准收益率29.74%
阿尔法0.31
贝塔0.94
夏普比率1.08
胜率0.51
盈亏比1.24
收益波动率31.33%
信息比率0.07
最大回撤25.39%

bigcharts-data-start/{"__type":"tabs","__id":"bigchart-97f63252fa66443b8796e66271405139"}/bigcharts-data-end