Hyperparameters

Overview#

The Hyperparameters panel provides a centralized interface for managing parameters used in optimization workflows. It supports automatic differentiation marking, allowing parameters to be optimized through gradient-based methods.

Panel Location#

The Hyperparameters panel appears as a left sidebar in the Node Editor. It can be toggled on/off from the Node Editor toolbar.

Parameter Types#

Adding Parameters#

1. Click [+ Add Parameter] button
2. Select parameter type from dropdown
3. Enter parameter name
4. Set initial value
5. Toggle differentiable if needed (∂ button)

Differentiable Toggle#

The ∂ button marks a parameter as differentiable:

• Enabled (blue): Parameter will be included in gradient computation
• Disabled (gray): Parameter is treated as constant

When a parameter is marked differentiable:

• Blue dashed animated border on the parameter
• Blue dashed animated edges in Node Editor
• Animation flows backward (indicating gradient direction)

Using Parameters in Node Editor#

Drag to Canvas#

Drag a parameter from the panel onto the Node Editor canvas:

1. Parameter node appears at drop location
2. Output port shows the parameter value
3. Connect to other nodes for computation

Node Connection Example#

┌─────────────────┐
│ learning_rate   │
│    Float: 0.001 │──→ [Multiply] ──→ [Update]
└─────────────────┘

Editing Parameters#

Inline Editing#

• Double-click name to rename
• Click value to edit
• Press Enter to confirm, Escape to cancel

Deleting Parameters#

• Right-click → Delete
• Or select and press Delete key
• Connected nodes are updated automatically

Sync Behavior#

Port Colors#

Parameter nodes follow the standard port color scheme:

Example: Optimization Setup#

Scene Optimization#

Hyperparameters:
├── tx_position (Vec3) [∂]  → Transmitter position
├── tx_power (Float) [∂]    → Transmitter power
├── frequency (Float) [ ]   → Fixed frequency
└── target_snr (Float) [ ]  → Target SNR (constant)
 
Node Graph:
tx_position → [Transmitter] → [Channel] → [SNR] → [Loss]
tx_power ──────────┘                              ↑
target_snr ──────────────────────────────────────┘

Material Optimization#

Hyperparameters:
├── roughness (Float) [∂]
├── metalness (Float) [∂]
└── color (Color) [∂]
 
Connect to Material component for real-time preview

Integration with Backend#

Parameters are synced with the backend for optimization:

# Backend receives parameter values and differentiable flags
# Gradient-based optimization can use marked parameters
 
# Parameters marked with ∂ are automatically tracked
# During optimization, gradients flow through connected nodes