Metamaze
Search…
Train and deploy
When there are sufficient documents labeled as training documents (at least 11 documents per language per document type) you can train or retrain the model.
There is a difference between training on project and document type level.
On document type level following trainings are possible:
  • Page management for this document type
  • Entity extraction for this document type
On project level following trainings are possible:
  • Page management for the project
  • Entity extraction deployment of training done on document type level
  • Document classification for document types added to the project
You can start a training for the document creation models (page management and/or document classification) and for the entity extraction models, per language. You can choose between training one multi-lingual model or one model per language.
You can always see which model has been rolled out and what the score was from the last training session. Based on the scores you can decide to roll out the new version of the model or to keep the current model.
In the overview you can see the total number of documents and the new documents. These new documents have not yet been used to train the model.
When you train the model with a new data set, you can compare the score with the current model.
And so you can decide whether there are additional documents to be handled or whether you want to roll out the new trained model.

Document classification

In the overview you can see the total number of documents and the new documents for each document type. These new documents have not yet been used to train the model.
By clicking on the 'Train now', a new model will be trained based on the current training data.
By clicking on the '+' sign, you get a historical overview of all trainings.
By clicking on the 'deploy'-button you can roll out the new or previous version of the model.
The overview shows the following data:
  • When the selected model has been trained
  • By whom the model was trained
  • How accurate the model is
  • When the model is rolled out for the last time
  • Whether the model has been deployed in production: "Deployed"
    • When you want to deploy another version, you have to click on 'deploy

Entity extraction

In this overview you can see the language, total number of documents and the new documents per type. These new documents have not yet been used to train the model. It also shows when this model was deployed and at which accuracy it is currently.
By clicking on the 'Train now', the following modal will open up. When the user selects 'train', a new model will be trained based on the current training data.
Besides choosing between a mono-lingual (1 language for instance English) and a multi-lingual (multiple languages) training, you can choose a full or incremental training. With the latter we refer to training a complete model or only training the model on new annotations that were added to the documents, this will impact the time needed for the model to train. An incremental training will drastically lower the training time.
By clicking on the '+' sign, you see a historical list of trained models.
By clicking on the 'deploy'-button in the column 'deployment' you can roll out the new or previous version of the model (only available in case you didn't deploy it yet).
The overview shows the following data:
  • When the selected model has been trained
  • By whom the model was trained
  • The model type
  • How accurate the model is
  • Whether the model has been deployed on the application : "Deployed"
When you want to deploy another version, you have to click on 'deploy'.
Last modified 1mo ago