为情感分析实验重新培训Web服务的问题 [英] Issues with retraining web service for Sentiment Analysis experiment

问题描述

我正在使用Azure ML Studio创建实验以执行情感分析，并尝试为此实验使用重新训练的Web服务，因此该模型可以通过编程方式进行训练.

在再培训服务中，我拥有词汇表和经过培训的模型作为Web服务输出，并将这些工件保存在Azure存储中.我更新了预测服务的资源(训练模型) 使用中提到的步骤 此处 .

In the retraining service, I have the vocabulary and the trained model as web service outputs and I save these artifacts in Azure Storage. I update the resource (trained model) for the prediction service using the steps mentioned here.

现在，当我测试预测服务时，出现以下错误:

 

Score Model (AFx Library) : table: The data set being scored must contain all features used during training, missing feature(s): 'Column 15629', 'Column 15630', 'Column 15631', 'Column 15632', 'Column 15633', 'Column 15634', 'Column 15635', 'Column 15636', 'Column 15637', 'Column 15638', 'Column 15639', 'Column 15640', 'Column 15641', 'Column 15642', 'Column 15643', 'Column 15644', 'Column 15645', 'Column 15646', 'Column 15647', 'Column 15648', 'Column 15649', 'Column 15650', 'Column 15651', 'Column 15652', 'Column 15653', 'Column 15654', 'Column 15655', 'Column 15656', 'Column 15657', 'Column 15658', 'Column 15659', 'Column 15660', 'Column 15661', 'Column 15662', 'Column 15663', 'Column 15664', 'Column 15665', 'Column 15666', 'Column 15667', 'Column 15668', 'Column 15669', 'Column 15670', 'Column 15671', 'Column 15672', 'Column 15673', 'Column 15674', 'Column 15675', 'Column 15676', 'Column 15677', 'Column 15678', 'Column 15679', 'Column 15680', 'Column 15681', 'Column 15682', 'Column 15683', 'Column 15684', 'Column 15685', 'Column 15686', 'Column 15687', 'Column 15688', 'Column 15689', 'Column 15690', 'Column 15691', 'Column 15692', 'Column 15693', 'Column 15694', 'Column 15695', 'Column 15696', 'Column 15697', 'Column 15698', 'Column 15699', 'Column 15700', 'Column 15701', 'Column 15702', 'Column 15703', 'Column 15704', 'Column 15705', 'Column 15706', 'Column 15707', 'Column 15708', 'Column 15709', 'Column 15710', 'Column 15711', 'Column 15712', 'Column 15713', 'Column 15714', 'Column 15715', 'Column 15716', 'Column 15717', 'Column 15718', 'Column 15719', 'Column 15720', 'Column 15721', 'Column 15722', 'Column 15723', 'Column 15724', 'Column 15725', 'Column 15726', 'Column 15727'. 

似乎如果您使用Web服务输出来保存经过训练的模型，则会添加一些额外的列(功能)

It seems like that if you use web service output to save the trained model, some extra columns (features) are added. 

还要注意，通常特征名称包括ngram，但是这些列名称没有ngram.

是否还有其他人在使用再培训服务来更新用于文本分析的模型?关于我可能做错了什么或如何解决此问题的任何建议?

推荐答案

当训练数据与计分数据具有不同的架构时，通常会显示此错误.同样，列名"Column< number>"也是如此.建议其中一个数据集缺少标题，因为这些看起来像是自动生成的无标题列名 数据集.

This error typically shows up when the training data has different schema than scoring data. Also column names "Column <number>" suggests that one of the datasets is missing a header, as these look like auto-generated column names for headerless datasets.

回到训练实验并仔细检查传递给训练模型"的列可能是有意义的.模块与计分实验中使用的模块匹配.

It probably makes sense to go back to the training experiment and double-check that the columns passed to "Train model" module match what's used in scoring experiment.

-Roope

这篇关于为情感分析实验重新培训Web服务的问题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！