MaterializationStrategy

Strategy interface for materializing tensor views into concrete tensors.

Materialization is the process of converting a tensor view (which references data from a parent tensor through coordinate transformations) into a standalone tensor with its own data storage. Different strategies provide different trade-offs between memory usage, computation time, and access patterns.

Strategy Patterns

• Immediate Materialization: Creates a copy of the view data immediately

• Lazy Materialization: Defers copying until the data is actually accessed

• Reference Materialization: Maintains references while optimizing access

• Batch Materialization: Optimizes for materializing multiple views together

Usage Context

Materialization strategies are particularly important for:

• Performance optimization when views are accessed frequently

• Memory management when parent tensors can be garbage collected

• Computational efficiency when the view will be used in multiple operations

• Interoperability when passing tensors between different computational backends

Parameters

Inheritors