年化26.8%，夏普1.28，单向做多螺纹钢的期货策略，Autogluon机器自主调参（策略+代码下载）。

继续“AI优化”的量化投资。

下面是Autogluon训练金融数据的基本流程：

from autogluon.core import TabularDataset
import numpy as np
from autogluon.tabular import TabularPredictor

from config import DATA_DIR
from engine.engine_utils import load_data

fields = ['shift(close,-1)/close -1', 'roc(close,20)', 'corr(close/shift(close,1), log(volume/shift(volume, 1)+1), 30)']
names = ['label_c', 'roc_20', 'CORR30']
symbols = ['510300.SH']

df = load_data(fields, names, symbols, columns=None, start_date='20100101',
               path=DATA_DIR.joinpath('etfs').resolve())

df['label'] = np.where(df['label_c'] > 0.0, 1, 0)
print(df['label'].value_counts())
del df['label_c']

split = int(len(df) * 0.8)
train = df.iloc[:split].copy()
test = df.iloc[split:].copy()
print('测试集日期', test.index[0])

label = 'label'
train_data = TabularDataset(train)
test_data = TabularDataset(test)

predictor = TabularPredictor(label=label, path='mymodel').fit(train_data)
print(predictor.leaderboard(test_data, silent=True))
y_pred = predictor.predict(test.drop(columns=[label]))
print(y_pred)
y_pred.to_csv('y_pred.csv')

训练集的准确率是：55.8%，目前我们仅引入了几个因子：

测试集上最好的模型是：LightGBMLarge，准确率是52.1%。

大家会觉得准确率不高，当然这里我们仅使用原始的价量数据是一方面，另外，胜率是一方面，还得看赔率。

我们使用期货——螺纹钢的历史日线数据来回测看下效果。

name = '机器择时'

[data]
start_date = '20100101'
test_start_date = '20210611'
end_date = ''
symbols = [
        # '510300.SH', # 沪深300ETF
         'RB0', # 螺纹钢
    ]
fields = ['shift(close,-1)/close -1','std(close, 20)/close','roc(close,20)','corr(close/shift(close,1), log(volume/shift(volume, 1)+1), 30)','corr(close/shift(close,1), log(volume/shift(volume, 1)+1), 30)']
names = ['label_c','std_20','roc_20','CORR30','CORR60']
data_folder = 'futures' # 数据在data下的目录
benchmark='RB0'

[model]
model_path='mymodel'
label=''
override=false

[[algos]]
name = 'SelectByModel'
model_path='mymodel'


[[algos]]
name = 'SelectBySignal'
rules=['buy']

[[algos]]
name = 'SelectBySignal'
rules=['sell']
exclude= true

[[algos]]
name = 'WeightEqually'

[[algos]]
name = 'Rebalance'