Concepts overview
This page offers an overview of what models and model versions are in Arkindex, and what they can be used for. For more details on specific features, see the advanced model documentation.
Models and model versions¶
A Model in Arkindex is a high-level object that represents the artifacts obtained by training a specific algorithm on specific data. Models can only be used in conjunction with workers by creating a Process. You can read the process documentation to find out more about what processes are and how to configure them.
For example, you might want to use a Named Entity Recognition worker on some documents. The worker itself takes your document images, elements, transcriptions etc, as input, and outputs entities depending on the selected model. If your documents are in French, you will select a model that has been trained on French data. Models can be further specialized: for example there might be a model that has been trained on French newspapers, or French census records.
However, just like with workers, you never actually use the model itself, but its Model versions. Model versions exist so that the models can improve and evolve, by being trained on additional data or with different parameters to obtain better results. When a new version of a model is trained, it is published on Arkindex, so that the latest version is always available. New model versions can also be published to follow changes in the workers they are used with, or in the libraries they use.
Info
If the data the model is trained on changes too much, it is preferable to create a new model instead of a new model version: the model object only makes sense if its model versions are suited to similar enough documents.
You can use Arkindex to train your own machine learning models.
Models on Arkindex¶
You can see a list of the models you have access to on an Arkindex instance by clicking Models in the user e-mail dropdown menu in the top-right corner.
You can see a list of a model’s model versions, as well as the workers it is compatible with, by clicking on its name in the list.
Compatible workers¶
Compatible workers can be defined for a model: this is a list of the workers this model is supposed to be used with. Users with administrator access to a model can manage its compatible workers from the model details, by clicking the Manage button on the right of the “Workers” heading.
When configuring a process and selecting a model version for a worker, by default only compatible models are displayed in the modal. However, you can always click the “Show all models” toggle to display all the models available to you on this instance. The list of compatible workers is not exclusive, and you can select a model version to go with a worker even if that worker is not amongst the compatible workers of the model; conversely, not all models have compatible workers defined.
Creating models and model versions¶
You can create models using the Arkindex frontend, by clicking the Create button above the list of available models. This opens a modal in which you must specify a name and description for your new model.
However, model versions can only be published using the Arkindex command line interface: see the model version publication command documentation.