liveBook · Manning

Model.parameters()

As explained in Deep Learning with PyTorch, Second Edition

A PyTorch module method that returns a module’s learnable parameters.

model.parameters() returns the learnable parameters owned by a PyTorch module and its registered submodules. A parameter is a tensor recognized by nn.Module and requiring gradients by default, so gradient descent can optimize it. section 6.2.2

The method recursively visits submodules assigned as attributes in the constructor and produces a flat sequence of their parameters. Passing that sequence to an optimizer connects backpropagated gradients with parameter updates. section 6.2.2 section 8.3.2

Passing parameters to SGD

linear_model = nn.Linear(1, 1)\noptimizer = optim.SGD(\n linear_model.parameters(),\n lr=1e-2)\n\nlist(linear_model.parameters())\n# [weight Parameter, bias Parameter] section 6.2.2

Keep these implementation details in mind:

  • Calling model.parameters() returns a generator; use list(model.parameters()) when you need to inspect the collected tensors. section 6.2.2
  • Register child modules as top-level attributes, or use nn.ModuleList and nn.ModuleDict for module collections; otherwise their parameters are not found. section 8.3.2
# In[11]:
linear_model.parameters()

# Out[11]:
<generator object Module.parameters at 0x7f94b4a8a750>

# In[12]:
list(linear_model.parameters())

# Out[12]:
[Parameter containing:
 tensor([[0.7398]], requires_grad=True), Parameter containing:
 tensor([0.7974], requires_grad=True)]

Inspect the generator and the weight and bias Parameters returned by a linear module. section 6.2.2

How Deep Learning with PyTorch, Second Edition covers it

How deep it goes
Explains recursive discovery, generator output, gradient accumulation, and optimizer updates, while stopping short of implementation internals.
Examples
Demonstrates nn.Linear weight and bias inspection, SGD construction, and parameter counting in a nested convolutional model.
What you can do after
Apply it via optim.SGD to train built-in or custom nn.Module models without passing parameters manually.
How it connects
Connects nn.Module, submodules, Parameter, autograd, gradient descent, optimizers, ModuleList, and ModuleDict.
Cover of Deep Learning with PyTorch, Second Edition
liveBookmanning.com
Read Deep Learning with PyTorch, Second Edition on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with PyTorch, Second Edition. The section links point to the parts of the book it draws on.