Graphically map out neural network structures

Table of Contents

• Quick Start
• Copyright and License

Quick Start

Option 1: Start with path to model

The first option is to start with some model state dict saved as either a .th or .pth file.

model = loadNetwork(file_path, architecture)

Note: A list of all supported network architectures can be printed out using the supportedNetworks() function found in utils.

Once we have loaded the model as shown above, it can then simply be processed and plotted using a call to the processNetwork() and plotNetwork() in utils as follows:

import utils

# Creating a dictionary of {name:weights} pairs
model_dict = utils.processNetwork(model)

# Plotting all trainable layers within the network using the dictionary we created in the previous line
utils.plotNetwork(model_dict)

Option 2: Start with preloaded model

If a model has already been created and either trained or loaded with a state dict, the model can then simply be processed and plotted using a call to the processNetwork() and plotNetwork() in utils. For example, let us say we have already loaded and trained some model named 'test_model". We would then process and plot the network with:

import utils

# Creating a dictionary of {name:weights} pairs
model_dict = utils.processNetwork(model)

# Plotting all trainable layers within the network using the dictionary we created in the previous line
utils.plotNetwork(model_dict)

Copyright and License