我正在关注 Notebook 作为如何使用Azure AutoML的示例,预测时间序列。
我有一个场景,我必须预测下个月不同商店和商品(大数据集)的销售额。每个月我都需要添加新数据并重新训练模型。
在AutoML实验运行之后,我将拥有最好的模型,它可用于进行预测,非常棒! / p>
但是现在我想重新训练那个模型,而不需要再次搜索最好的模型。
我该怎么做?
解决方案
你好,
您可以下载最佳运行的模型文件,并使用它来创建 azure ML管道
为了获得最佳的运行迭代,你可以 检查结果并从门户网站迭代的输出选项卡下载模型文件。
您还可以通过在AutoMLConfig中将enable_ensembling的bool更改为True并使用此迭代模型进行新数据集的训练来进行集成。
Hi,
I am following this Notebook as an example on how to use Azure AutoML, to forecast a time series.
I have a scenario where I must forecast next month sales for different stores and items (large data sets). And every month I need to add new data and retrain the models.
After the AutoML experiment run, I will have the best model and it can be used to make predictions, great!.
But now I want to retrain that model without the need to search for the best model all over again.
How can I do that?
解决方案Hello,
You can download the model file of the run that is best and use it to create a azure ML pipeline
To get the best iteration of a run you can examinethe results and download the model file from output tab of the iteration in portal.
You can also enable ensembling by changing the bool of enable_ensembling to True in AutoMLConfig and use this iteration model for training with new datasets.
更多推荐
重新训练AutoML最佳模型
发布评论