Amazon Forecast 不再向新买家开放。Amazon Forecast 的现有客户可以继续照常使用该服务。[了解更多](https://aws.amazon.com/blogs/machine-learning/transition-your-amazon-forecast-usage-to-amazon-sagemaker-canvas/)

本文属于机器翻译版本。若本译文内容与英语原文存在差异，则一律以英文原文为准。

# Amazon Forecast 算法
<a name="aws-forecast-choosing-recipes"></a>

Amazon Forecast 预测器使用算法通过时间序列数据集训练模型。然后，使用经过训练的模型来生成指标和预测。

 如果您不确定要使用哪种算法来训练模型，请在创建预测器时选择 AutoML，然后让 Forecast 为您的数据集训练最优模型。或者，您可以手动选择其中一种 Amazon Forecast 算法。

**Python 笔记本**  
有关使用 AutoML 的 step-by-step指南，请参阅 AutoML [入门](https://github.com/aws-samples/amazon-forecast-samples/blob/master/notebooks/advanced/Getting_started_with_AutoML/Getting_started_with_AutoML.ipynb)。

## 内置 Forecast 算法
<a name="forecast-algos"></a>

 Amazon Forecast 提供了六种内置算法供您选择。这些算法包括自回归积分滑动平均模型（ARIMA）等常用统计算法，以及 CNN-QR 和 DeepAR\+ 等复杂的神经网络算法。

### [CNN-QR](aws-forecast-algo-cnnqr.md)
<a name="cnnqr"></a>

 `arn:aws:forecast:::algorithm/CNN-QR` 

 Amazon Forecast CNN-QR，卷积神经网络——分位数回归，是一种专有的机器学习算法，用于使用因果卷积神经网络预测时间序列 ()。CNNsCNN-QR 最适合处理包含数百个时间序列的大型数据集。它接受项目元数据，并且是唯一接受不包含未来值的相关时间序列数据的 Forecast 算法。

### [DeepAR\+](aws-forecast-recipe-deeparplus.md)
<a name="deeparplus"></a>

`arn:aws:forecast:::algorithm/Deep_AR_Plus`

 Amazon Forecast Deepar\+ 是一种专有的机器学习算法，用于使用循环神经网络预测时间序列 () RNNs。DeepAR\+ 最适合处理包含数百个特征时间序列的大型数据集。该算法接受前瞻性相关时间序列和项目元数据。

### [Prophet](aws-forecast-recipe-prophet.md)
<a name="prophet"></a>

`arn:aws:forecast:::algorithm/Prophet`

 Prophet 是一种基于加性模型的时间序列预测算法，其中非线性趋势与每年、每周和每日的季节性相吻合。它最适合具有强季节效应的时间序列和多个季节的历史数据。

### [NPTS](aws-forecast-recipe-npts.md)
<a name="npts"></a>

`arn:aws:forecast:::algorithm/NPTS`

 Amazon Forecast 非参数时间序列 (NPTS) 专有算法是可扩展的概率基线预测器。NPTS 在处理稀疏或间歇性时间序列时特别有用。Forecast 提供了四种算法变体：标准 NPTS、季节性 NPTS、气候学预报器和季节性气候学预报器。

### [ARIMA](aws-forecast-recipe-arima.md)
<a name="arima"></a>

`arn:aws:forecast:::algorithm/ARIMA`

 自回归积分滑动平均模型 (ARIMA) 是一种常用的时间序列预测统计算法。该算法对小于 100 个时间序列的简单数据集特别有用。

### [ETS](aws-forecast-recipe-ets.md)
<a name="ets"></a>

`arn:aws:forecast:::algorithm/ETS`

 指数平滑法 (ETS) 是一种常用的时间序列预测统计算法。该算法对小于 100 个时间序列的简单数据集以及具有季节性模式的数据集特别有用。ETS 计算时间序列数据集中所有观察数据的加权平均值作为其预测，权重随时间呈指数递减。

## 比较 Forecast 算法
<a name="comparing-algos"></a>

 使用下表查找最适合您的时间序列数据集的选项。


<table>
<thead>
  <tr><th></th><th colspan="2">神经网络</th><th>灵活的局部算法</th><th colspan="3">基线算法</th></tr>
  <tr><th></th><th>CNN-QR</th><th>DeepAR+</th><th>Prophet</th><th>NPTS</th><th>ARIMA</th><th>ETS</th></tr>
</thead>
<tbody>
  <tr><td>计算密集型训练过程</td><td>高</td><td>高</td><td>中</td><td>低</td><td>低</td><td>低</td></tr>
  <tr><td>接受历史相关时间序列*</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>接受前瞻性相关时间序列*</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>接受项目元数据（商品颜色、品牌等）</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>接受天气指数内置特征化</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>适用于稀疏数据集</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>执行超参数优化（HPO）</td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
  <tr><td>允许覆盖默认的超参数值 </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-yes.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td><td><img src="http://docs.aws.amazon.com/zh_cn/forecast/latest/dg/images/icon-no.png" alt="" /> </td></tr>
</tbody>
</table>


\*有关相关时间序列的更多信息，请参阅[相关时间序列](related-time-series-datasets.md)。