agviz

AutoGrad Visualizer (agviz): Visualize dependencies between states/parameters in autograd tree.

Unlike related packages (e.g. torvhviz), supports visualizing internal model states and distinguishing states/parameters. Also supports more complex visualization based on traversals of graph and merging nodes.

Installation:

In the root directory run pip install -e .
You can then install the library in your python code using `from adviz import render' (see Examples)

Examples:

examples/gru: Render the hidden states of a GRU for 5 timesteps, visualizing parameters and states.

examples/rnn: Render the hidden states of a RNN for 5 timesteps, visualizing parameters and states. Also, play with "full" rendering, where we show all the intermediate functions.