Skip to main content
from onyxengine.modeling import Output

feature = Output(
    name: str,
    dataset_feature: Optional[str] = None,
    scale: Union[None, Literal['mean'], List[float]] = 'mean',
    parent: Optional[str] = None,
    relation: Optional[Literal['equal', 'delta', 'derivative']] = None
)
Defines an output feature for the model - what the neural network predicts.

​
Parameters

​
name
str
required
Name of the output feature. Used for dictionary access in simulation results.
​
dataset_feature
str
default:"None"
Column name in the dataset. If None, uses name.
​
scale
Union[None, 'mean', List[float]]
default:"mean"
Scaling method:
  • 'mean': Normalize to zero mean, unit std (default)
  • None: No scaling
  • [min, max]: Scale from range to [-1, 1]
​
parent
str
default:"None"
Parent output name for derived outputs. Required if relation is set.
​
relation
Literal
default:"None"
Relationship to parent for derived outputs:
  • 'derivative': output[t+1] = output[t] + parent[t] * dt
  • 'delta': output[t+1] = output[t] + parent[t]
  • 'equal': output[t+1] = parent[t]
  • None: Direct prediction (default)

​
Output Types

​
Direct Output

Most outputs are direct predictions from the neural network: 
Output(name='acceleration')
Output(name='torque')
Output(name='force')

​
Derived Output

Outputs can derive from other outputs (less common): 
outputs = [
    Output(name='jerk'),  # Neural network predicts jerk
    Output(name='acceleration', parent='jerk', relation='derivative'),  # Derived
]

​
Examples

​
Single Output

from onyxengine.modeling import Output

outputs = [Output(name='acceleration')]

​
Multiple Outputs

outputs = [
    Output(name='acceleration_x'),
    Output(name='acceleration_y'),
    Output(name='acceleration_z'),
]

​
Custom Dataset Column Name

# Feature named 'acceleration' maps to 'imu_accel_x' in dataset
Output(name='acceleration', dataset_feature='imu_accel_x')

​
Custom Scaling

# Scale acceleration from [-50, 50] m/s² to [-1, 1]
Output(name='acceleration', scale=[-50, 50])

# No scaling
Output(name='normalized_output', scale=None)

​
Derived Outputs

# Neural network predicts jerk, acceleration is derived
outputs = [
    Output(name='jerk'),
    Output(name='acceleration', parent='jerk', relation='derivative'),
]

​
Simulation Access

After simulation, access outputs by name: 
result = model.simulate(x0=..., external_inputs=..., sim_steps=100)

# Access specific output
acceleration = result.outputs['acceleration']  # Shape: (batch, sim_steps, 1)

# Access all outputs as tensor
all_outputs = result.outputs.tensor  # Shape: (batch, sim_steps, num_outputs)

​
Validation

  • If relation is set, parent must also be set
  • If parent is set, relation must also be set
  • For derived outputs, parent must be another Output name