search

google-research / FirstOrderLp.jl

Experimental first-order solvers for linear and quadratic programming.
Apache License 2.0
103 stars 19 forks source link

Diagonal scaling is slow for large instances and may raise an out of memory error #118

Closed haihaolu closed 1 year ago

haihaolu commented 1 year ago

Expected Behavior

Diagonal scaling should be very fast.

Actual Behavior

Diagonal scaling is slow for large instances and may raise an out of memory error.

Steps to Reproduce the Problem

  1. Use the code with default parameters to solve a MIPLIB instance bab3 may raise an out of memory error.
  2. The reason for this comes from line 570-573 in preprocess.jl. "Diagonal(1 ./ constraint_rescaling) problem.constraint_matrix Diagonal(1 ./ variable_rescaling)" creates dense matrices. This may be due to the new Julia version.

Specifications