Tutorials

• CSRNDArray - NDArray in Compressed Sparse Row Storage Format
  • Advantages of Compressed Sparse Row NDArray (CSRNDArray)
  • Prerequisites
  • Compressed Sparse Row Matrix
    • Example Matrix Compression
  • Array Creation
  • Inspecting Arrays
  • Storage Type Conversion
  • Copies
  • Indexing and Slicing
  • Sparse Operators and Storage Type Inference
  • Data Loading
  • Advanced Topics
    • GPU Support
  • Next
• RowSparseNDArray - NDArray for Sparse Gradient Updates
  • Motivation
  • Prerequisites
  • Row Sparse Format
  • Array Creation
  • Function Overview
  • Setting Type
  • Inspecting Arrays
  • Storage Type Conversion
  • Copies
  • Retain Row Slices
  • Sparse Operators and Storage Type Inference
  • Sparse Optimizers
  • Advanced Topics
    • GPU Support
  • Next
• Train a Linear Regression Model with Sparse Symbols
  • Prerequisites
  • Variables
    • Variable Storage Types
    • Bind with Sparse Arrays
  • Symbol Composition and Storage Type Inference
    • Basic Symbol Composition
    • Storage Type Inference
    • Storage Type Fallback
    • Inspecting Storage Types of the Symbol Graph
  • Training with Module APIs
    • Preparing the Data
    • Defining the Model
    • Training the model
    • Training the model with multiple machines or multiple devices
• Sparse NDArrays with Gluon
  • Generating Sparse Data
  • Writing Sparse Data
  • Reading Sparse Data
    • Gluon Models for Sparse Data
  • Benchmark: FullyConnected
  • Benchmark: FullyConnectedSparse
  • Benchmark: FullyConnectedSparse with grad_stype=row_sparse
  • Advanced: Sparse weight
    • Conclusion
    • Recommended Next Steps