nn.Flatten
We initialize the nn.Flatten layer to convert each 2D 28x28 image into a contiguous array of 784 pixel values ( the minibatch dimension (at dim=0) is maintained)
nn.Sequential
nn.Sequential is an ordered container of modules. The data is passed through all the modules in the same order as defined. You can use sequential containers to put together a quick network like seq_modules
.
nn.Softmax
The last linear layer of the neural network returns logits - raw values in [-infty, infty] - which are passed to the nn.Softmax module. The logits are scaled to values [0, 1] representing the model’s predicted probabilities for each class. dim
parameter indicates the dimension along which the values must sum to 1.
Model Parameters
Many layers inside a neural network are parameterized, i.e. have associated weights and biases that are optimized during training. Subclassing nn.Module
automatically tracks all fields defined inside your model object, and makes all parameters accessible using your model’s parameters()
or named_parameters()
methods.
AUTOMATIC DIFFERENTIATION WITH TORCH.AUTOGRAD
When training neural networks, the most frequently used algorithm is back propagation. In this algorithm, parameters (model weights) are adjusted according to the gradient of the loss function with respect to the given parameter.
To compute those gradients, PyTorch has a built-in differentiation engine called torch.autograd
. It supports automatic computation of gradient for any computational graph.
Consider the simplest one-layer neural network, with input x
, parameters w
and b
, and some loss function. It can be defined in PyTorch in the following manner:
In this network, w
and b
are parameters, which we need to optimize. Thus, we need to be able to compute the gradients of loss function with respect to those variables. In order to do that, we set the requires_grad
property of those tensors.
NOTE
You can set the value of requires_grad
when creating a tensor, or later by using x.requires_grad_(True)
method.
A function that we apply to tensors to construct computational graph is in fact an object of class Function
. This object knows how to compute the function in the forward direction, and also how to compute its derivative during the backward propagation step. A reference to the backward propagation function is stored in grad_fn
property of a tensor. You can find more information of Function
in the documentation.
- We can only obtain the
grad
properties for the leaf nodes of the computational graph, which haverequires_grad
property set toTrue
. For all other nodes in our graph, gradients will not be available. - We can only perform gradient calculations using
backward
once on a given graph, for performance reasons. If we need to do severalbackward
calls on the same graph, we need to passretain_graph=True
to thebackward
call.
By default, all tensors with requires_grad=True
are tracking their computational history and support gradient computation. However, there are some cases when we do not need to do that, for example, when we have trained the model and just want to apply it to some input data, i.e. we only want to do forward computations through the network. We can stop tracking computations by surrounding our computation code with torch.no_grad()
block:
with torch.no_grad():
z = torch.matmul(x, w)+b
print(z.requires_grad)
Another way to achieve the same result is to use the detach()
method on the tensor:
There are reasons you might want to disable gradient tracking:
- To mark some parameters in your neural network as frozen parameters. This is a very common scenario for finetuning a pretrained network
- To speed up computations when you are only doing forward pass, because computations on tensors that do not track gradients would be more efficient.
Conceptually, autograd keeps a record of data (tensors) and all executed operations (along with the resulting new tensors) in a directed acyclic graph (DAG) consisting of Function objects. In this DAG, leaves are the input tensors, roots are the output tensors. By tracing this graph from roots to leaves, you can automatically compute the gradients using the chain rule.
In a forward pass, autograd does two things simultaneously:
- run the requested operation to compute a resulting tensor
- maintain the operation’s gradient function in the DAG.
The backward pass kicks off when .backward()
is called on the DAG root. autograd
then:
- computes the gradients from each
.grad_fn
, - accumulates them in the respective tensor’s
.grad
attribute - using the chain rule, propagates all the way to the leaf tensors