分数模型与交叉验证的分数数据集输出 [英] Scored datasets output from Score Model vs. Cross Validate
问题描述
你好
根据经验，最好在整个标记的数据集而不是子集上训练预测模型.我是在Python中以这种方式进行的.唯一的问题是评估模型.交叉验证是一种解决方法， 因为该模型是在多个内部折叠中进行评估的，这些折叠被用作迭代测试数据集.我试图避免将数据集拆分为训练和测试子集，因为这将减少可用于训练模型的数据量.
As a rule of thumb, it is better to train a predictive model on an entire labelled dataset rather than a subset of it. I do it this way in Python. The only problem with that is in evaluating the model. Cross-validation is a way around that, as the model is evaluated over multiple, internal folds that are used as iterative test datasets. I seek to avoid Splitting the dataset into Train and Test subsets, as this will reduce the amount of data available to train the model.
我正在AzureML中构建一个模型，以制作一个Web服务，该服务将为未标记的数据集生成预测的标签.我正在大量标签数据上训练模型.
I am building a model in AzureML to make a Web Service that will generate predicted labels for an unlabeled dataset. I am training the model on a large set of labelled data.
在这种情况下，最好这样做:
In such a case, is it better to:
A.使用Tune Model Hyperparameters基于整个数据集(未提供验证数据集)生成经过训练的模型，将该模型输出到Cross Validate 由 交叉验证生成，作为Web的最终输出服务，或
A. Use Tune Model Hyperparameters to generate a trained model based on an entire dataset (no validation dataset provided), output that model to Cross Validate along with the entire dataset, and use the Scored Dataset generated by Cross Validate as final output for the Web Service, or
B. ，将该模型连同整个数据集一起输出到得分模型，并且 使用 由评分模型
B. Use Tune Model Hyperparameters to generate a trained model based on an entire dataset (no validation dataset provided), output that model to Score Model along with the entire dataset, and use the Scored Dataset generated by Score Model as final output for the Web Service?
我最困惑的是交叉验证模块的计分数据集"输出.有没有人有使用该输出的任何示例?我从未见过.这意味着什么?
The main thing I am confused about is the Scored Dataset output of the Cross Validate module. Does anyone have any examples of that output being used? I have never seen one. What is it meant to be used for?
谢谢...
推荐答案
请遵循以下文档来创建预测性Web服务.
Please follow the following document for creating a predictive web service.
加上更多评论:
- 典型做法是将整个数据集分为训练，测试和评估子集.训练和测试子集用于训练模型.评估子集仅用于一次从各种训练模型中选择最终一个.
- 调整模型超参数和交叉验证模块旨在进行模型训练和选择
- 要实现此目标，请遵循上述创建Web服务的准则.
这篇关于分数模型与交叉验证的分数数据集输出的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！
