Class NodeClassificationTask<T>

Namespace

AiDotNet.Data.Structures

Assembly

AiDotNet.dll

Represents a node classification task where the goal is to predict labels for individual nodes.

public class NodeClassificationTask<T>

Type Parameters

T

The numeric type used for calculations, typically float or double.

Inheritance

object

NodeClassificationTask<T>

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Remarks

Node classification is a fundamental graph learning task where each node in a graph has a label, and the goal is to predict labels for unlabeled nodes based on: - Node features - Graph structure (connections between nodes) - Labels of neighboring nodes

For Beginners: Node classification is like categorizing people in a social network.

Real-world examples:

Social Networks:

• Nodes: Users
• Task: Predict user interests/communities
• How: Use profile features + friend connections
• Example: "Is this user interested in sports?"

Citation Networks:

• Nodes: Research papers
• Task: Classify paper topics
• How: Use paper abstracts + citation links
• Example: Papers citing each other often share topics

Fraud Detection:

• Nodes: Financial accounts
• Task: Detect fraudulent accounts
• How: Use transaction patterns + account relationships
• Example: Fraudsters often form connected clusters

Key Insight: Node classification leverages the graph structure. Connected nodes often share similar properties (homophily), so a node's neighbors provide valuable information for prediction.

Properties

Graph

The graph data containing nodes, edges, and features.

public GraphData<T> Graph { get; set; }

Property Value

GraphData<T>

Remarks

This is the complete graph structure. In semi-supervised node classification, some nodes have known labels (training set) and others don't (test set).

IsMultiLabel

Whether this is a multi-label classification task.

public bool IsMultiLabel { get; set; }

Property Value

bool

Remarks

• False: Each node has exactly one label (e.g., paper topic)
• True: Each node can have multiple labels (e.g., user interests)

Labels

Node labels for all nodes in the graph. Shape: [num_nodes] for single-label or [num_nodes, num_classes] for multi-label.

public Tensor<T> Labels { get; set; }

Property Value

Tensor<T>

Remarks

In semi-supervised settings, labels for test nodes are only used for evaluation, not during training.

NumClasses

Number of classes in the classification task.

public int NumClasses { get; set; }

Property Value

int

TestIndices

Indices of nodes to use for testing.

public int[] TestIndices { get; set; }

Property Value

int[]

TrainIndices

Indices of nodes to use for training.

public int[] TrainIndices { get; set; }

Property Value

int[]

Remarks

For Beginners: In semi-supervised node classification, we typically have: - Small set of labeled nodes for training (5-20% of nodes) - Larger set of unlabeled nodes for testing

This split simulates real-world scenarios where getting labels is expensive. For example, manually labeling research papers by topic requires expert knowledge.

ValIndices

Indices of nodes to use for validation.

public int[] ValIndices { get; set; }

Property Value

int[]