Chainer Reference Manual

• Core functionalities
  • Variable
  • Flag
  • Function
  • Link and Chain
  • Optimizer
  • Serializer
  • Dataset abstraction
  • Training loop abstraction
  • Debug mode
  • FunctionSet (deprecated)
• Utilities
  • CUDA utilities
  • Common algorithms
  • Reporter
  • Experimental feature annotation
• Assertion and Testing
  • Type checking utilities
  • Gradient checking utilities
  • Standard Assertions
  • Function testing utilities
• Standard Function implementations
  • Activation functions
  • Array manipulations
  • Neural network connections
  • Evaluation functions
  • Loss functions
  • Mathematical functions
  • Noise injections
  • Normalization functions
  • Spatial pooling
  • Utility functions
• Standard Link implementations
  • Learnable connections
  • Activation/loss/normalization functions with parameters
  • Machine learning models
  • Pre-trained models
  • Deprecated links
• Optimizers
• Serializers
  • Serialization in NumPy NPZ format
  • Serialization in HDF5 format
• Function hooks
  • Base class
  • Concrete function hooks
• Weight Initializers
  • Base class
  • Concrete initializers
  • Helper function
• Dataset examples
  • General datasets
  • Concrete datasets
• Iterator examples
  • SerialIterator
  • MultiprocessIterator
• Trainer extensions
  • dump_graph
  • Evaluator
  • ExponentialShift
  • LinearShift
  • LogReport
  • observe_lr
  • observe_value
  • snapshot
  • snapshot_object
  • PlotReport
  • PrintReport
  • ProgressBar
• Trainer triggers
  • Interval
  • ManualSchedule
  • Minimum and maximum values
• Caffe Reference Model Support
• Visualization of Computational Graph
• Environment variables