[Coding] nn.Linear

nn.Linear is used to create a linear transform layer.

 

Parameters

• in_features, out_features: The shape of the input and output.
• bias(True): If false the bias is removed.
• dtype(None): Sets the dtype of tensors inside the layer. If set to 'None' it will default to 'torch.float32'

A lab is provided, but here are some important parts…

• The linear transformation is made up of a weight and bias matrix.
• The shape of the weight matrix is (out, in), but when it is multiplied, it is changed to (in, out).
• The shape of the bias matrix is (out), but if the bias option is False then it doesn’t exist.
• The dtype of the tensor is set to torch.float32 if not specified.
• You can use model.weight and model.bias to get the actual matrices.

NNLinearModel.ipynb

0.02MB