Decipher: can not Load Batch when Training

[sumire]

I referred to this page:
https://docs.blueprism.com/ja-JP/bundle/decipher-idp-2-3/page/user-guide/classification-model-training-guide.htm

• "C:\Program Files (x86)\Blue Prism\Decipher Automated Clients\SsiDataCaptureClient.exe.config"
  set "EnableModelTrainingML" to True
  
• create Capture Models
  set "Mark for Training" to ON
  
• create Classification Models
  set "Marked for training" to OFF, "Extensible" to ON
  
• amend Document Types
  set "Machine Learning" to ON, in "ML Model" select , set "Training Size" to 100
  
• amend Batch Types
  set "Classification mode" to Semantic
  in "Classification model" select the created Classification Model name
  
• create process to push a batch of training documents
  
• check the batch status is "Completed Class Training"
  

• Load batch into class verification
  ...my batch doesn't appear!
  

Are there any errors or omissions in the work procedure I have written?

Please help me

[Ben.Lyons1]

Hi,

Classification training batches are only used for training the classification model. They will not appear for verification, nor will they contribute towards training the Capture ML model (that's for data extraction only).

Once a classification training batch has completed it's training, it can be deleted.

Also, you don't need to create a Capture ML model or assign a capture model to the document type in order to perform the classification model training, or even to use the classification model after training. They're entirely separate functions.

Thanks

[sumire]

Hello @Ben.Lyons1 ,

Thank you for your reply.
But I don't understand about classification model training.
I followed the instructions on this page:
https://docs.blueprism.com/ja-JP/bundle/decipher-idp-2-3/page/user-guide/classification-model-training-guide.htm 
Am I referring to the wrong page?
I use 1 batch type with 7 document types, and would like to be able to classify them correctly.
I would also like to improve the reading accuracy of characters (Japanese), but I recognize that I should refer to another page other than the above URL.

[Ben.Lyons1 · Accepted solution]

Hi @sumire ,

That's the correct page.

Classification training batches are only used to create/train the classification model. They do not pass through data extraction and will never appear in data verification.

This is because a file uploaded for data capture may contain multiple documents and the classification model will separate them. The classification model training files must already be separated into the documents, so the model can accurately be trained on what the document type looks like.

Have you completed the Decipher foundation training?

If this is still unclear, I would recommend either raising a support ticket or contacting your Blue Prism account manager on how to get additional support.

The OCR engine cannot be trained, so character extraction cannot be improved. It could be that the document quality is not sufficient.

Decipher 2.3.2 will include the ability to use one of Azure's OCR engines (for additional cost), which may provide more accurate OCR capture.

Thanks

[sumire]

Hello @Ben.Lyons1 ,

Thank you for your explanation.
Since Japanese OCR accuracy is low, I will look into Azure's OCR engines.