Table of Contents

Class GraphClassificationTask<T>

Namespace
AiDotNet.Data.Structures
Assembly
AiDotNet.dll

Represents a graph classification task where the goal is to classify entire graphs.

public class GraphClassificationTask<T>

Type Parameters

T

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

Inheritance
GraphClassificationTask<T>
Inherited Members

Remarks

Graph classification assigns a label to an entire graph based on its structure and node/edge features. Unlike node classification (classify individual nodes) or link prediction (predict edges), graph classification treats the whole graph as a single data point.

For Beginners: Graph classification is like determining the category of a complex object.

Real-world examples:

Molecular Property Prediction:

  • Input: Molecular graph (atoms as nodes, bonds as edges)
  • Task: Predict molecular properties
  • Examples:
    • Is this molecule toxic?
    • What is the solubility?
    • Will this be a good drug candidate?
  • Dataset: ZINC, QM9, BACE

Protein Function Prediction:

  • Input: Protein structure graph
  • Task: Predict protein function or family
  • How: Analyze amino acid sequences and 3D structure

Chemical Reaction Prediction:

  • Input: Reaction graph showing reactants and products
  • Task: Predict reaction type or outcome

Social Network Analysis:

  • Input: Community subgraphs
  • Task: Classify community type or behavior
  • Example: Identify bot networks vs organic communities

Code Analysis:

  • Input: Abstract syntax tree (AST) or control flow graph
  • Task: Detect bugs, classify code functionality
  • Example: "Is this code snippet vulnerable to SQL injection?"

Key Challenge: Graph-level representation

  • Must aggregate information from all nodes and edges
  • Common approaches: Global pooling, hierarchical pooling, set2set

Properties

AvgNumEdges

Average number of edges per graph (for informational purposes).

public double AvgNumEdges { get; set; }

Property Value

double

AvgNumNodes

Average number of nodes per graph (for informational purposes).

public double AvgNumNodes { get; set; }

Property Value

double

IsMultiLabel

Whether this is a multi-label classification task.

public bool IsMultiLabel { get; set; }

Property Value

bool

Remarks

  • False: Each graph has exactly one label (e.g., molecule is toxic or not)
  • True: Each graph can have multiple labels (e.g., molecule has multiple properties)

IsRegression

Whether this is a regression task instead of classification.

public bool IsRegression { get; set; }

Property Value

bool

Remarks

For regression tasks (e.g., predicting molecular energy), labels are continuous values rather than discrete classes.

For Beginners: The difference between classification and regression: - **Classification**: Predict categories (e.g., "toxic" vs "non-toxic") - **Regression**: Predict continuous values (e.g., "solubility = 2.3 mg/L")

Examples:

  • Classification: Is this molecule a good drug? (Yes/No)
  • Regression: What is this molecule's binding affinity? (0.0 to 10.0)

NumClasses

Number of classes in the classification task.

public int NumClasses { get; set; }

Property Value

int

TestGraphs

List of test graphs.

public List<GraphData<T>> TestGraphs { get; set; }

Property Value

List<GraphData<T>>

TestLabels

Labels for test graphs. Shape: [num_test_graphs] or [num_test_graphs, num_classes].

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

Property Value

Tensor<T>

TrainGraphs

List of training graphs.

public List<GraphData<T>> TrainGraphs { get; set; }

Property Value

List<GraphData<T>>

Remarks

Each graph in the list is an independent sample with its own structure and features.

TrainLabels

Labels for training graphs. Shape: [num_train_graphs] or [num_train_graphs, num_classes] for multi-label.

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

Property Value

Tensor<T>

ValGraphs

List of validation graphs.

public List<GraphData<T>> ValGraphs { get; set; }

Property Value

List<GraphData<T>>

ValLabels

Labels for validation graphs. Shape: [num_val_graphs] or [num_val_graphs, num_classes].

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

Property Value

Tensor<T>
Edit this page