Namespace MOSAIC.Models.Learning
Classes
HybridPredictorBlock
Hybrid Predictor Block - combines classifier training with continuous vector output.
Training: Like a classifier - learns from discrete class labels (action names from Trigger) Output: Continuous vector - weighted sum of class target vectors by predicted probabilities
Example:
Classes defined in Trigger:
rest: [0,0,0,0,0,0]
power: [1,1,1,1,0,0]
pinch: [0,0,0,0,1,1]
Predicted probabilities: [0.1, 0.7, 0.2]
Output = 0.1rest + 0.7power + 0.2*pinch = [0.7, 0.7, 0.7, 0.7, 0.2, 0.2]
This gives smooth, continuous control signals for prosthesis control while training with simple discrete gesture labels.
Inputs:
- Input[0]: Data source (feature vectors)
- Input[1]: Trigger (provides class labels AND target vectors)
JSON Config: { "Type": "HybridPredictor", "Inputs": ["DataSource", "Trigger"], "DesiredRate": 200, "Params": ["SoftmaxRFF", 8, 0.01, 0.001, 1.0, 300], "Path": "models/" }
Params: [modelType, inputDim, learningRate?, lambda?, sigma?, featureDim?] Note: Classes are automatically learned from Trigger actions!
HybridPredictorState
State for serializing HybridPredictor.