eqiora.torch

← Python API index

Use Eqiora differentiable programs with PyTorch autograd.

View source

Example: differentiate a scalar objective

Install the torch extra, then create program with the differentiation example:

python

import torch
from eqiora.torch import bind

operator = bind(program)
parameters = torch.tensor([1.5], dtype=torch.float64, requires_grad=True)
values = operator(parameters)
values.square().sum().backward()
print(parameters.grad)

The gradient is the derivative of the sum of squared output values with respect to the selected parameter. Inputs are contiguous, one-dimensional CPU float64 tensors. See framework adapters for installation and compiled execution.

API

TorchProgram

Functional PyTorch view of one Eqiora program, valid in this process.

python

class TorchProgram:
    def __init__(self, program: DifferentiableProgram) -> None: ...
    @property
    def program(self) -> DifferentiableProgram: ...
    @property
    def input_shape(self) -> tuple[int]: ...
    @property
    def output_shape(self) -> tuple[int]: ...
    def __call__(self, parameters: torch.Tensor) -> torch.Tensor: ...

bind

Register a program as a PyTorch operator in this process.

python

def bind(program: DifferentiableProgram) -> TorchProgram: ...