DyNet Neural Network Toolkit

This is going to provide a repository of basic documentation for using the DyNet library.

To work with the DyNet library, you will want to understand the following concepts:

• Models (which collections of parameters)
• Parameters belong to models. They are either Matrices/Vectors representing e.g., weight matrices or bias vectors and the like. Or they are vectors obtained from a lookup table (e.g. get me the embedding of the word "happy").
• Computation graphs represented as a hypergraph where nodes are the results of a computation--always a vector or matrix and edges link the arguments going into a function with the label that the function produces. Importantly, these computation graphs make use of parameters and compute functions -- ultimately objective functions that will be optimized.
• Symbolic Variables are nodes in the computation graph.