DocsNode Editor

Node Editor

The visual heart of CyxWiz Engine, allowing you to design neural network architectures through an intuitive drag-and-drop interface.

Overview

80+ Node Types

Comprehensive ML building blocks

Visual Connections

Type-safe data flow links

Shape Inference

Automatic dimension calculation

Code Generation

Export to PyTorch, TensorFlow, Keras

Patterns Library

Pre-built architecture templates

Undo/Redo

Full history support

Interface

+------------------------------------------------------------------+
|  [Toolbar: Save | Load | Generate Code | Validate | Train]       |
+------------------------------------------------------------------+
|                                                                   |
|     +----------+        +----------+        +----------+          |
|     | DataInput|------->| Dense    |------->| ReLU     |          |
|     | [MNIST]  |        | units:128|        |          |          |
|     +----------+        +----------+        +----------+          |
|                              |                   |                |
|                              |              +----------+          |
|                              +------------->| Dense    |          |
|                                             | units:10 |          |
|                                             +----------+          |
|                                                  |                |
|                                             +----------+          |
|                                             | Softmax  |          |
|                                             +----------+          |
|                                                  |                |
|                                             +----------+          |
|                                             | Output   |          |
|                                             +----------+          |
|                                                                   |
|                                           [Minimap]               |
+------------------------------------------------------------------+

Node Categories

Data Pipeline

Entry points and data handling:

Dataset InputData LoaderAugmentationData SplitNormalizeOne-Hot Encode

Core Layers

Fundamental neural network layers:

DenseConv1D/2D/3DMaxPool2DAvgPool2DGlobalMaxPoolAdaptiveAvgPool

Recurrent

Sequence processing:

RNNLSTMGRUBidirectionalTimeDistributedEmbedding

Attention & Transformer

Modern attention mechanisms:

MultiHeadAttentionSelfAttentionCrossAttentionTransformerEncoderTransformerDecoderPositionalEncoding

Activations

Non-linear functions:

ReLULeakyReLUGELUSwishSigmoidTanhSoftmax

Normalization

Batch and layer normalization:

BatchNormLayerNormGroupNormInstanceNorm

Optimizers

Training algorithms:

SGDAdamAdamWRMSpropAdagradNAdam

Loss Functions

Training objectives:

MSELossCrossEntropyLossBCELossL1LossHuberLoss

Pin Types

Nodes connect via typed pins:

Type	Color	Purpose
Tensor	Blue	General tensor data
Labels	Green	Classification labels
Parameters	Orange	Model parameters
Loss	Red	Loss values
Optimizer	Purple	Optimizer state
Dataset	Cyan	Dataset reference

Navigation

Mouse Controls

Left-click	Select node
Ctrl+Left-click	Add to selection
Left-drag (empty)	Box select
Middle-drag	Pan canvas
Scroll wheel	Zoom
Right-click	Context menu

Keyboard Shortcuts

F	Frame selected
Home	Frame all
Delete	Delete selected
Ctrl+D	Duplicate
Ctrl+A	Select all
Ctrl+F	Search nodes

Tips & Best Practices

Start with Data - Always begin with DatasetInput
End with Output - Required for training
Validate Often - Catch errors early
Use Patterns - Start from proven architectures
Name Your Nodes - Double-click title to rename
Organize Spatially - Use alignment tools
Save Frequently - Use auto-save