TensorBlock#

using metatensor_torch::TorchTensorBlock = torch::intrusive_ptr<TensorBlockHolder>#

TorchScript will always manipulate TensorBlockHolder through a torch::intrusive_ptr

class TensorBlockHolder : public CustomClassHolder#

Wrapper around metatensor::TensorBlock for integration with TorchScript

Python/TorchScript code will typically manipulate torch::intrusive_ptr<TensorBlockHolder> (i.e. TorchTensorBlock) instead of instances of TensorBlockHolder.

Public Functions

TensorBlockHolder(torch::Tensor data, TorchLabels samples, std::vector<TorchLabels> components, TorchLabels properties)#

Create a new TensorBlockHolder with the given data and metadata.

TensorBlockHolder(metatensor::TensorBlock block, torch::IValue parent)#

Create a torch TensorBlockHolder from a pre-existing metatensor::TensorBlock.

If the block is a view inside another TorchTensorBlock or TorchTensorMap, then parent should point to the corresponding object, making sure a reference to it is kept around.

TorchTensorBlock copy() const#

Make a copy of this TensorBlockHolder, including all the data contained inside

torch::Tensor values()#

Get a view in the values in this block.

TorchLabels labels(uintptr_t axis) const#

Get the labels in this block associated with either "values" or one gradient (by setting values_gradients to the gradient parameter); in the given axis.

inline TorchLabels samples() const#

Access the sample Labels for this block.

The entries in these labels describe the first dimension of the values() array.

inline std::vector<TorchLabels> components() const#

Access the component Labels for this block.

The entries in these labels describe intermediate dimensions of the values() array.

inline TorchLabels properties() const#

Access the property Labels for this block.

The entries in these labels describe the last dimension of the values() array. The properties are guaranteed to be the same for values and gradients in the same block.

void add_gradient(const std::string &parameter, TorchTensorBlock gradient)#

Add a set of gradients with respect to parameters in this block.

Parameters:
  • parameter – add gradients with respect to this parameter (e.g. "positions", "cell", …)

  • gradient – a TorchTensorBlock whose values contain the gradients with respect to the parameter. The labels of the gradient TorchTensorBlock should be organized as follows: its samples must contain "sample" as the first label, which establishes a correspondence with the samples of the original TorchTensorBlock; its components must contain at least the same components as the original TorchTensorBlock, with any additional component coming before those; its properties must match those of the original TorchTensorBlock.

inline std::vector<std::string> gradients_list() const#

Get a list of all gradients defined in this block.

bool has_gradient(const std::string &parameter) const#

Check if a given gradient is defined in this TensorBlock.

std::string repr() const#

Implementation of repr/__str__ for Python.

inline const metatensor::TensorBlock &as_metatensor() const#

Get the underlying metatensor TensorBlock.

Public Static Functions

static TorchTensorBlock gradient(TorchTensorBlock self, const std::string &parameter)#

Get a gradient from this TensorBlock.

static std::vector<std::tuple<std::string, TorchTensorBlock>> gradients(TorchTensorBlock self)#

Get a all gradients and associated parameters in this block.