Now that we have some data labelled, let's start using bootstrapping models to speed up our annotation.
Creating a model
We will start by creating a new model by going to the 'Models' tab in the project and clicking 'Create model'.
Let's a put a title and description on the model, so we can keep track of it later.
Now we can select options for instantiating this new model. The first and most important is what we want this model to do under 'Model type': 'Classification', 'Object detection', or 'Instance segmentation'.
Framework and model
We can also select the machine learning framework we want the model to use under 'Framework', and the specific model class under 'Model'.
Classes and objects
Finally, we want to select the classes or objects we want the model to register. For object detection and segmentation this is the specific objects to include. For classification, it is which "classification question" we want the model to "answer". This can be any question within the classification tree in your ontology.
Training a model
We have now instantiated the model with what we want it to do and how he want it to do it. We can now go ahead and train the model with some labelled data. First, click on 'Train'.
We can set the number of epochs we want the model to train for. Under 'Advanced settings' we can additionally set the 'Batch size' if it is trained with a CPU or GPU (GPU recommended) and what set of training weights to initialize the model with.
Now we can select what data to include to train the model. Select the data you want to use, click 'Train' and now let the learning begin!
Tracking progress status
You can keep track of the progress of the model training by expanding the snack bar arrow at the top of the screen.
Once the training is complete, the status of the model will update to say it is 'Trained'.