thecosta / networkviewer

Visualization tool for feed-forward neural networks with Python's Matplotlib.

NetworkViewer

Visualization tool for feed-forward networks.

Directions for use

Example scripts can be found in the examples folder.

Each layer of your network should be created with the Layer object.

Layer Attributes

• name: layer name. must include 'dense', 'conv', 'input', or 'output' if layer represents one of such.
• output_dim: dimensionality of the output layer. must be Tuple class.
• depth: depth or number of channels of the output layer.
• output_dim_label: output dimensionality label to use. useful for when reshaping output dimensions. tuple in this repository are represented by class Tuple, with attributes x and y.

Your network needs to be translated into a list of Layer. feed them into the Plotter class.

Plotter Attributes

• layers: list of Layer.
• window_width: pixel width of plotting window.
• window_height: pixel height of plotting window.
• dpi: dots per inch.
• padding: pixel padding between layers.
• show_dim: show layer dimensions.
• show_lower_text: show lower layer text, ie activation functions maxpooling, flatten.
• title: plot title.
• x_scale: scale all x dimensions.
• y_scale: scale all y dimensions.
• depth_scale: scale all depth dimensions.
• log: scale by log factor all dimensions.

Examples

Baseline Model

VGGNet

AlexNet