- 微信
- 微博
  
  分享文章到微博
- 复制链接
  
  复制链接到剪贴板

用DeepAR做股票价格预测

darkpard 发表于 2022/09/12 09:39:04 2022/09/12

【摘要】 DeepAR是亚马逊提出的一种时间序列预测算法，通过在大量时间序列上训练自回归递归网络模型，从相关的时间序列中有效学习全局模型，进而对时间序列进行预测。它能应对季节性、周期性等问题。本文探索下该模型在股票预测中的应用。安装包!pip install baostock==0.8.8!pip install mxnet==1.7.0.post2!pip install gluonts==0.9....

DeepAR是亚马逊提出的一种时间序列预测算法，通过在大量时间序列上训练自回归递归网络模型，从相关的时间序列中有效学习全局模型，进而对时间序列进行预测。它能应对季节性、周期性等问题。本文探索下该模型在股票预测中的应用。

安装包

!pip install baostock==0.8.8
!pip install mxnet==1.7.0.post2
!pip install gluonts==0.9.5
!pip install matplotlib==3.3.1

2. 导入包

import baostock as bs
from gluonts.dataset import common
from gluonts.model import deepar
from gluonts.mx.trainer import Trainer
from gluonts.evaluation.backtest import make_evaluation_predictions
from tqdm.autonotebook import tqdm
import os
import matplotlib.pyplot as plt

3. 数据准备

3.1. 变量初始化

stock_list = ['sh.603198', 'sh.600809', 'sh.600519'] #以三个白酒股为例
train_list = []
test_list = []
prediction_length = 20 #预测长度

3.2. 数据下载函数

def get_stockdata(code):
    rs = bs.query_history_k_data_plus(code,
                                      'date,close,volume,turn',
                                      start_date='2020-09-01',
                                      end_date='2022-09-01',
                                      frequency='d', adjustflag='2') 
    return rs.get_data()

3.3. 连上baostock

lg = bs.login()

3.4. 数据下载与划分

for stock_id in stock_list:
    df = get_stockdata(stock_id)
    train_dic = {'start':df.date[0],
                 'target':df.close,
                 'cat':int(stock_id.split('.')[1]),
                 'dynamic_feat':[df.volume, df.turn]}
        test_dic = {'start':df.date[0],
                'target':df.close[:-prediction_length],
                'cat':int(stock_id.split('.')[1]),
                'dynamic_feat':[df.volume[:-prediction_length], df.turn[:-prediction_length]]}
    train_list.append(train_dic)
    test_list.append(test_dic)

3.5. 数据封装

train_data = common.ListDataset(train_list, freq='1d')
test_data = common.ListDataset(test_list, freq='1d')

4. 建模与训练

4.1. 构建训练器

estimator = deepar.DeepAREstimator(
    prediction_length=prediction_length,
    context_length=60,
    freq='1d',
    num_layers=2,
    num_cells=64,
    trainer=Trainer(epochs=20,
                    learning_rate=1e-2,
                    num_batches_per_epoch=32))

4.2. 模型训练

predictor = estimator.train(train_data)

5. 股价预测

forecast_it, ts_it = make_evaluation_predictions(
    dataset=test_data,
    predictor=predictor,
    num_samples=100)

6. 模型评估与可视

6.1. 数据封装

tss = list(tqdm(ts_it, total=len(test_data)))
forecasts = list(tqdm(forecast_it, total=len(test_data)))

6.2. 新建存放文件夹

plot_log_path = './plots/'
directory = os.path.dirname(plot_log_path)
if not os.path.exists(directory):
    os.makedirs(directory)

6.3. 结果可视化

def plot_prob_forecasts(ts_entry, forecast_entry, path, sample_id):
    plot_length = 150
    prediction_intervals = (50, 80)
    legend = ['observations', 'median prediction'] + [f'{k}% prediction interval' for k in prediction_intervals][::-1]
    _, ax = plt.subplots(1, 1, figsize=(10, 7))
    ts_entry[-plot_length:].plot(ax=ax)
    forecast_entry.plot(prediction_intervals=prediction_intervals, color='g')
    ax.axvline(ts_entry.index[-prediction_length], color='r')
    plt.legend(legend, loc='upper left')
    plt.savefig('{}forecast_{}.png'.format(path, sample_id))
    plt.close()

6.4. 将结果存入刚新建的文件夹

for i in tqdm(range(len(stock_list))):
    ts_entry = tss[i]
    forecast_entry = forecasts[i]
    plot_prob_forecasts(ts_entry, forecast_entry, plot_log_path, i)