Utils

Introduction

Utils are general utilties used by other modules in Zenkai.

Key Utilities

The utilities are used by the core modules, Tansaku, Kikai to make it easier to implement the framework. They are as follows:

• Parameter - There are several utilities for retrieving and setting parameters or there gradients

• Sampling - There are several utilities for retrieving and setting parameters or there gradients

• Parameter Utilities:

  from zenkai.utils import get_model_parameters, set_model_parameters, get_model_grads, set_model_grads
  
  model = nn.Linear(2, 2)
  # this wraps parameters_to_vector to make it a little simpler
  parameters = get_model_parameters(model)
  parameters = parameters * 2
  set_model_parameters(model, parameters)
  
  # use to retrieve the grads from the model for storing etc
  grads = get_model_grads(model)
  grads = grads * 0.1
  # can update them as well
  set_model_grads(model)
  
  # another somewhat related function is freshen
  x = torch.randn(4, 2)
  x = model(x)
  # this will detach it and set x to retain grads
  # this is necessary in the LearningMachine framework to prevent
  # grads from being backpropagated
  freshen(x)