Unreal Engine has many asset types to build experiences and games with such as meshes, textures, sounds, music, animations, etc. These are objects that the application can call upon (loaded into memory) when needed to build and run the world the user experiences. InteractML has it's own custom asset types to help manage the training examples and model state used by running models. These are as follows, with more detail below.

Classification A trained classification algorithm	Regression A trained regression algorithm	Dynamic Timewarp A trained dynamic timewarp algorithm
Training Set Examples a specific model was trained with	Label Definition of a composite Label, the type of an Expected Output or Output	Label Table List of Expected Output labels each with explicit values assigned

Assets are added to your project via the InteractML section of the Unreal Content Browser context menu.

The three model types have their own corresponding model asset. These store the infomation to operate the model independantly of the examples it was trained with. It is these assets that you train and run using the corresponding "Teach The Machine" and "Machine Learning Robot" Blueprint nodes.

See Model Types for more details 👉
See Training Models & Running Models for model use 👉
See Your Data for how model data is stored 👉

The training set asset stores recorded input parameters and the expected outputs that go with each one. This infomation is applied to an appropriate model to train it using the "Teach The Machine" Blueprint node.

See Training Set for more details 👉
See Recording Examples for how to build training sets 👉
See Your Data for how example data is stored 👉

The label asset is used to define composite label types. These define the various values that make up a model output or expected ouput when recording examples. They consist of a list of named types (much like the fields in a structure). These allow the use of multidimensional outputs, particularly useful with the continuous interpolation operation of the Regression model.

See Labels for more details 👉
See Training Models & Running Models for label use 👉

As part of the training process it is sometimes useful to be able to pre-define the sets of values that correspond to specific output configurations desired. By using the Label Asset definitions we can build such a table that can be simply indexed when selecting the output we are training for. This helps simplify the training script, reduce chance of error when selecting output to train for, and also speeds up the training process as it is much simpler to select/switch between them.

See Labels for more details 👉
See Training Models for label table use 👉

---

👈 Model Types | 🏠 Home | Blueprint Nodes 👉