Create a model
The models you create will be stored on Decthings server and automatically executed when needed. To create a model, go to the create model page. Which configurations options to choose depends on your use case.
- Train an already existing model on your own dataset
- Use the visual editor
- Create a model by writing code
Train an already existing model on your own dataset
If the dataset that a pre-made model was trained on is of little relevance for your application you can re-train the model to better fit your needs. For example, a generic image classifier will be able to classify images in a broad range of classes such as "cat", "dog" and "tennis ball", but your application might be more interested in classifying types of birds. In this case, you can provide your own dataset of birds and re-train the model.
For this option, select executor "Based on existing" in the create model dialog. This will create a model which does not contain any code of its own, but instead is based on a snapshot of another model. The benefit of this is that you can re-use a single model for many different applications. For example, you can create a model which can classify different types of animals by selecting a generic image classifier as a base and then retraining that model on your own dataset of animals. This way, you can create a specific image classifier without writing the code yourself.
You can also use this option if you want to customize some options for the model, such as the LauncherSpec or the maximum durations for operations, but don't want to modify the original model.
Important: Because a snapshot cannot be modified, you can be assured that your created model stays the same even if the original model gets edited. However, if the original model or the original snapshot gets deleted, any future operations (evaluations, training, etc.) will fail to start.
Create a new model from scratch
You can also create your own model from scratch, either by using the visual editor or by writing code. The visual editor allows you to create many types of models such as image classifiers, text-to-speech synthesizers or image generators. While the visual editor aims to cover as many use cases as possible, one can never really beat the flexibility of writing your own code.
