CVXTorch

This package generates Pytorch expressions from CVXPY expressions and constraints. A Pytorch expression is a function that evaluates similarly to a CVXPY's numeric function. A torch expression and a mapping from CVXPY leaves to their indices in the Pytorch expression are returned.

Installation

To install CVXTorch, clone the repo from the source, and

pip install -e .

Example

import cvxpy as cp
import torch
from cvxtorch import TorchExpression

n = 5
x = cp.Variable(n, name="x")
y = cp.Parameter(n, name="y")
z = 3
exp = x-y+2*z

tch_x = torch.arange(1, n+1)
tch_y = torch.arange(0, n)

tch_exp = TorchExpression(exp).torch_expression #tch_exp implements x-y+2*z, where x and y are torch.Tensor.
tch_res = tch_exp(tch_x, tch_y) #Contains a torch.Tensor [7.0]*n

Returned object

A CVXTorch TorchExpression object is created. It has two properties:

• torch_expression (callable): The generated torch expression.

• variables_dictionary (cvxtorch.VariablesDict): An object that maps from CVXPY atoms to their indices in the generated torch expression.