TorchLean API

Docs HomeInstallationGuideExamplesGraphsTools

NN.Examples.DeepDives.TensorOperations

Tensor Operations #

This example continues after NN.Examples.Quickstart.TensorBasics. It shows indexing, einsum, batch axes, reshaping, elementwise operations, and editor widgets on the same shape-indexed tensor type used by models and specifications.

Build with lake build NN.Examples.DeepDives.TensorOperations, then inspect the #tensor_view commands in the editor. This module has no command-line entry point.

Construction and indexing #

source
def NN.Examples.DeepDives.TensorOperations.matrix :
TorchLean.Tensor Float [2, 3]

A matrix literal. Its element type and both dimensions are visible in the type.

Instances For
    source
    def NN.Examples.DeepDives.TensorOperations.firstRow :
    TorchLean.Tensor Float [3]

    Selecting one coordinate on the leading axis returns a tensor of the remaining shape.

    Instances For

      Linear algebra #

      source
      def NN.Examples.DeepDives.TensorOperations.weights :
      TorchLean.Tensor Float [4, 3]

      A 4 × 3 parameter tensor.

      Instances For
        source
        def NN.Examples.DeepDives.TensorOperations.input :
        TorchLean.Tensor Float [3]

        A three-component input vector.

        Instances For
          source
          def NN.Examples.DeepDives.TensorOperations.linearOutput :
          TorchLean.Tensor Float [4]

          Matrix multiplication preserves the output dimension in the result type.

          Instances For

            Batches and reshaping #

            source
            def NN.Examples.DeepDives.TensorOperations.batch :
            TorchLean.Tensor Float [2, 2, 3]

            Two 2 × 3 samples stacked along a leading batch axis.

            Instances For
              source
              def NN.Examples.DeepDives.TensorOperations.firstSample :
              TorchLean.Tensor Float [2, 3]

              Indexing a rank-three tensor drops the outer axis, giving a [2, 3] matrix.

              Instances For
                source
                def NN.Examples.DeepDives.TensorOperations.secondSample :
                TorchLean.Tensor Float [2, 3]

                The second batch element, same shape.

                Instances For
                  source
                  def NN.Examples.DeepDives.TensorOperations.flatValues :
                  TorchLean.Tensor Float [6]

                  A flat tensor whose values will be viewed at another shape.

                  Instances For
                    source
                    def NN.Examples.DeepDives.TensorOperations.reshapedMatrix :
                    TorchLean.Tensor Float [2, 3]

                    Reshape a vector without changing row-major element order.

                    Instances For

                      Shape-preserving transformations #

                      source
                      def NN.Examples.DeepDives.TensorOperations.tensorA :
                      TorchLean.Tensor Float [2, 2]

                      A constant matrix of twos for the elementwise addition example.

                      Instances For
                        source
                        def NN.Examples.DeepDives.TensorOperations.tensorB :
                        TorchLean.Tensor Float [2, 2]

                        A constant matrix of threes; adding it to tensorA produces a matrix of fives.

                        Instances For
                          source
                          def NN.Examples.DeepDives.TensorOperations.elementwiseSum :
                          TorchLean.Tensor Float [2, 2]

                          Ordinary + on tensors is elementwise, and the shapes must already agree.

                          There is no implicit broadcasting here: a shape mismatch is a type error, not a runtime surprise.

                          Instances For
                            source
                            def NN.Examples.DeepDives.TensorOperations.doubledOutput :
                            TorchLean.Tensor Float [4]

                            A pointwise transformation has the same shape discipline as a gradient buffer.

                            Instances For