深度学习特征列表稍微做修改就不行了，咋回事？

albertech · 2020-08-19 04:35:30 UTC

克隆策略

In [2]:

# 本代码由可视化策略环境自动生成 2020年8月19日 12:32
# 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 m4_run_bigquant_run(input_1, input_2, input_3):
    # 示例代码如下。在这里编写您的代码
    df =  input_1.read_pickle()
    feature_len = len(input_2.read_pickle())
    
    
    df['x'] = df['x'].reshape(df['x'].shape[0], int(feature_len), int(df['x'].shape[1]/feature_len))
    
    data_1 = DataSource.write_pickle(df)
    return Outputs(data_1=data_1)

# 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 m4_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 m8_run_bigquant_run(input_1, input_2, input_3):
    # 示例代码如下。在这里编写您的代码
    df =  input_1.read_pickle()
    feature_len = len(input_2.read_pickle())
    
    
    df['x'] = df['x'].reshape(df['x'].shape[0], int(feature_len), int(df['x'].shape[1]/feature_len))
    
    data_1 = DataSource.write_pickle(df)
    return Outputs(data_1=data_1)

# 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 m8_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 m24_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[:,0], '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 m24_post_run_bigquant_run(outputs):
    return outputs

# 回测引擎：初始化函数，只执行一次
def m19_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=0.0003, sell_cost=0.0013, 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'] = 2

# 回测引擎：每日数据处理函数，每天执行一次
def m19_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 m19_prepare_bigquant_run(context):
    pass


m1 = M.instruments.v2(
    start_date='2014-01-01',
    end_date='2015-01-01',
    market='CN_STOCK_A',
    instrument_list=' ',
    max_count=0
)

m2 = M.advanced_auto_labeler.v2(
    instruments=m1.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, -2) / 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
)

m13 = M.standardlize.v8(
    input_1=m2.data,
    columns_input='label'
)

m3 = M.input_features.v1(
    features="""close_0/mean(close_0,5)
close_0/mean(close_0,10)
close_0/mean(close_0,20)
close_0/open_0
open_0/mean(close_0,5)
open_0/mean(close_0,10)
open_0/mean(close_0,20)
(high_0-open_0/2-close_0/2)/(close_0-open_0)
(high_1-open_1/2-close_1/2)/(close_1-open_1)"""
)

m15 = M.general_feature_extractor.v7(
    instruments=m1.data,
    features=m3.data,
    start_date='',
    end_date='',
    before_start_days=0
)

m16 = M.derived_feature_extractor.v3(
    input_data=m15.data,
    features=m3.data,
    date_col='date',
    instrument_col='instrument',
    drop_na=True,
    remove_extra_columns=False
)

m14 = M.standardlize.v8(
    input_1=m16.data,
    input_2=m3.data,
    columns_input=''
)

m7 = M.join.v3(
    data1=m13.data,
    data2=m14.data,
    on='date,instrument',
    how='inner',
    sort=False
)

m26 = M.dl_convert_to_bin.v2(
    input_data=m7.data,
    features=m3.data,
    window_size=5,
    feature_clip=5,
    flatten=True,
    window_along_col='instrument'
)

m4 = M.cached.v3(
    input_1=m26.data,
    input_2=m3.data,
    run=m4_run_bigquant_run,
    post_run=m4_post_run_bigquant_run,
    input_ports='',
    params='{}',
    output_ports=''
)

m9 = M.instruments.v2(
    start_date=T.live_run_param('trading_date', '2015-01-01'),
    end_date=T.live_run_param('trading_date', '2017-01-01'),
    market='CN_STOCK_A',
    instrument_list='',
    max_count=0
)

m17 = M.general_feature_extractor.v7(
    instruments=m9.data,
    features=m3.data,
    start_date='',
    end_date='',
    before_start_days=0
)

m18 = M.derived_feature_extractor.v3(
    input_data=m17.data,
    features=m3.data,
    date_col='date',
    instrument_col='instrument',
    drop_na=True,
    remove_extra_columns=False
)

m25 = M.standardlize.v8(
    input_1=m18.data,
    input_2=m3.data,
    columns_input=''
)

m27 = M.dl_convert_to_bin.v2(
    input_data=m25.data,
    features=m3.data,
    window_size=5,
    feature_clip=5,
    flatten=True,
    window_along_col='instrument'
)

m8 = M.cached.v3(
    input_1=m27.data,
    input_2=m3.data,
    run=m8_run_bigquant_run,
    post_run=m8_post_run_bigquant_run,
    input_ports='',
    params='{}',
    output_ports=''
)

m6 = M.dl_layer_input.v1(
    shape='9,5',
    batch_shape='',
    dtype='float32',
    sparse=False,
    name=''
)

m10 = M.dl_layer_conv1d.v1(
    inputs=m6.data,
    filters=20,
    kernel_size='3',
    strides='1',
    padding='valid',
    dilation_rate=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=''
)

m12 = M.dl_layer_maxpooling1d.v1(
    inputs=m10.data,
    pool_size=1,
    padding='valid',
    name=''
)

m32 = M.dl_layer_conv1d.v1(
    inputs=m12.data,
    filters=20,
    kernel_size='3',
    strides='1',
    padding='valid',
    dilation_rate=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=''
)

m33 = M.dl_layer_maxpooling1d.v1(
    inputs=m32.data,
    pool_size=1,
    padding='valid',
    name=''
)

m28 = M.dl_layer_globalmaxpooling1d.v1(
    inputs=m33.data,
    name=''
)

m30 = M.dl_layer_dense.v1(
    inputs=m28.data,
    units=1,
    activation='linear',
    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=''
)

m34 = M.dl_model_init.v1(
    inputs=m6.data,
    outputs=m30.data
)

m5 = M.dl_model_train.v1(
    input_model=m34.data,
    training_data=m4.data_1,
    optimizer='RMSprop',
    loss='mean_squared_error',
    metrics='mae',
    batch_size=256,
    epochs=5,
    n_gpus=0,
    verbose='2:每个epoch输出一行记录'
)

m11 = M.dl_model_predict.v1(
    trained_model=m5.data,
    input_data=m8.data_1,
    batch_size=1024,
    n_gpus=0,
    verbose='2:每个epoch输出一行记录'
)

m24 = M.cached.v3(
    input_1=m11.data,
    input_2=m18.data,
    run=m24_run_bigquant_run,
    post_run=m24_post_run_bigquant_run,
    input_ports='',
    params='{}',
    output_ports=''
)

m19 = M.trade.v4(
    instruments=m9.data,
    options_data=m24.data_1,
    start_date='',
    end_date='',
    initialize=m19_initialize_bigquant_run,
    handle_data=m19_handle_data_bigquant_run,
    prepare=m19_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'
)

[2020-08-19 12:30:51.212290] INFO: moduleinvoker: instruments.v2 开始运行..

[2020-08-19 12:30:51.226999] INFO: moduleinvoker: 命中缓存

[2020-08-19 12:30:51.228828] INFO: moduleinvoker: instruments.v2 运行完成[0.016516s].

[2020-08-19 12:30:51.231580] INFO: moduleinvoker: advanced_auto_labeler.v2 开始运行..

[2020-08-19 12:30:51.240642] INFO: moduleinvoker: 命中缓存

[2020-08-19 12:30:51.241976] INFO: moduleinvoker: advanced_auto_labeler.v2 运行完成[0.010393s].

[2020-08-19 12:30:51.243958] INFO: moduleinvoker: standardlize.v8 开始运行..

[2020-08-19 12:30:51.250991] INFO: moduleinvoker: 命中缓存

[2020-08-19 12:30:51.252090] INFO: moduleinvoker: standardlize.v8 运行完成[0.008127s].

[2020-08-19 12:30:51.253403] INFO: moduleinvoker: input_features.v1 开始运行..

[2020-08-19 12:30:51.257537] INFO: moduleinvoker: 命中缓存

[2020-08-19 12:30:51.258721] INFO: moduleinvoker: input_features.v1 运行完成[0.005307s].

[2020-08-19 12:30:51.264891] INFO: moduleinvoker: general_feature_extractor.v7 开始运行..

[2020-08-19 12:30:52.196112] INFO: 基础特征抽取: 年份 2014, 特征行数=569948

[2020-08-19 12:30:52.452832] INFO: 基础特征抽取: 年份 2015, 特征行数=0

[2020-08-19 12:30:53.084547] INFO: 基础特征抽取: 总行数: 569948

[2020-08-19 12:30:53.088322] INFO: moduleinvoker: general_feature_extractor.v7 运行完成[1.823424s].

[2020-08-19 12:30:53.089988] INFO: moduleinvoker: derived_feature_extractor.v3 开始运行..

[2020-08-19 12:30:54.224965] INFO: derived_feature_extractor: 提取完成 close_0/mean(close_0,5), 0.984s

[2020-08-19 12:30:55.181096] INFO: derived_feature_extractor: 提取完成 close_0/mean(close_0,10), 0.955s

[2020-08-19 12:30:56.326902] INFO: derived_feature_extractor: 提取完成 close_0/mean(close_0,20), 1.145s

[2020-08-19 12:30:56.329633] INFO: derived_feature_extractor: 提取完成 close_0/open_0, 0.002s

[2020-08-19 12:30:57.269804] INFO: derived_feature_extractor: 提取完成 open_0/mean(close_0,5), 0.939s

[2020-08-19 12:30:58.273998] INFO: derived_feature_extractor: 提取完成 open_0/mean(close_0,10), 1.003s

[2020-08-19 12:30:59.202625] INFO: derived_feature_extractor: 提取完成 open_0/mean(close_0,20), 0.927s

[2020-08-19 12:30:59.205653] INFO: derived_feature_extractor: 提取完成 (high_0-open_0/2-close_0/2)/(close_0-open_0), 0.002s

[2020-08-19 12:30:59.208343] INFO: derived_feature_extractor: 提取完成 (high_1-open_1/2-close_1/2)/(close_1-open_1), 0.002s

[2020-08-19 12:30:59.374875] INFO: derived_feature_extractor: /y_2014, 569948

[2020-08-19 12:31:00.234703] INFO: moduleinvoker: derived_feature_extractor.v3 运行完成[7.14469s].

[2020-08-19 12:31:00.236524] INFO: moduleinvoker: standardlize.v8 开始运行..

[2020-08-19 12:31:02.123091] INFO: moduleinvoker: standardlize.v8 运行完成[1.88654s].

[2020-08-19 12:31:02.125121] INFO: moduleinvoker: join.v3 开始运行..

[2020-08-19 12:31:03.036585] INFO: join: /data, 行数=512393/518020, 耗时=0.732907s

[2020-08-19 12:31:03.463536] INFO: join: 最终行数: 512393

[2020-08-19 12:31:03.478724] INFO: moduleinvoker: join.v3 运行完成[1.353595s].

[2020-08-19 12:31:03.484220] INFO: moduleinvoker: dl_convert_to_bin.v2 开始运行..

[2020-08-19 12:31:23.055907] INFO: moduleinvoker: dl_convert_to_bin.v2 运行完成[19.571653s].

[2020-08-19 12:31:23.059632] INFO: moduleinvoker: cached.v3 开始运行..

[2020-08-19 12:31:24.933728] INFO: moduleinvoker: cached.v3 运行完成[1.87408s].

[2020-08-19 12:31:24.938482] INFO: moduleinvoker: dl_layer_input.v1 运行完成[0.002801s].

[2020-08-19 12:31:24.957744] INFO: moduleinvoker: dl_layer_conv1d.v1 运行完成[0.017559s].

[2020-08-19 12:31:24.966270] INFO: moduleinvoker: dl_layer_maxpooling1d.v1 运行完成[0.006644s].

[2020-08-19 12:31:24.986592] INFO: moduleinvoker: dl_layer_conv1d.v1 运行完成[0.018161s].

[2020-08-19 12:31:24.994112] INFO: moduleinvoker: dl_layer_maxpooling1d.v1 运行完成[0.005692s].

[2020-08-19 12:31:24.999418] INFO: moduleinvoker: dl_layer_globalmaxpooling1d.v1 运行完成[0.003535s].

[2020-08-19 12:31:25.012234] INFO: moduleinvoker: dl_layer_dense.v1 运行完成[0.011104s].

[2020-08-19 12:31:25.028911] INFO: moduleinvoker: cached.v3 开始运行..

[2020-08-19 12:31:25.452808] INFO: moduleinvoker: cached.v3 运行完成[0.423867s].

[2020-08-19 12:31:25.454155] INFO: moduleinvoker: dl_model_init.v1 运行完成[0.440285s].

[2020-08-19 12:31:25.455845] INFO: moduleinvoker: dl_model_train.v1 开始运行..

[2020-08-19 12:31:25.577635] INFO: device_manager: 没有gpu资源，将使用cpu计算

[2020-08-19 12:31:25.580974] INFO: device_manager: 本次操作不使用GPU

[2020-08-19 12:31:25.912558] INFO: dl_model_train: 准备训练，训练样本个数：512393，迭代次数：5

[2020-08-19 12:31:59.591029] INFO: dl_model_train: 训练结束，耗时：33.68s

[2020-08-19 12:32:02.459443] INFO: moduleinvoker: dl_model_train.v1 运行完成[37.003572s].

Train on 512393 samples
Epoch 1/5
512393/512393 - 8s - loss: nan - mae: nan
Epoch 2/5
512393/512393 - 7s - loss: nan - mae: nan
Epoch 3/5
512393/512393 - 7s - loss: nan - mae: nan
Epoch 4/5
512393/512393 - 6s - loss: nan - mae: nan
Epoch 5/5
512393/512393 - 6s - loss: nan - mae: nan

训练时报错：
loss: nan - mae: nan

请给位高手指点迷津。
谢谢！

tiantianz · 2020-09-16 02:58:52 UTC

要不您刷新一下再试试？我这边是可以运行出来的耶。

albertech · 2020-09-19 08:26:11 UTC

请注意：loss:nan-mae:nan
这样的是不行的。你那边不是这样的结果？请截图贴出来晒一下。

adhaha111 · 2020-09-22 05:43:51 UTC

您好，您的策略中没有 nan 处理模块，导致传入网络的数据中存在nan，这样训练出来的loss就会是nan了，您可以在python自定义模块中处理下nan再试试呢