High-performance Open-source Mixed-Integer Linear Optimization Solver (HiGHS) Python Interface

Warnings

• gotcha Direct iteration or element-wise access of returned array-like solution values (e.g., `solution.col_value`) can be very slow. It is highly recommended to convert these to Python lists first for efficient access.
  Fix: Convert solution arrays to lists: `col_value = list(solution.col_value)` before accessing elements.
• gotcha When constructing linear expressions, the `highs_linear_expression.__add__` method modifies the expression in-place. Reusing an expression object after it has been modified can lead to unexpected and hard-to-debug results.
  Fix: If an expression needs to be reused, ensure you are working with a copy or construct expressions explicitly for each use. More robust expression building might be available in newer versions (e.g., `ExprBuilder` mentioned in related discussions).
• gotcha By default, HiGHS C++ logging is duplicated to `Highs.log`. In `highspy`, to redirect logging output to a specific file, you must explicitly set the 'log_file' option.
  Fix: Use `h.setOptionValue('log_file', 'your_log_file.txt')` on your `highspy.Highs` instance to specify a log file.
• gotcha Users integrating `highspy` with other modeling libraries like `python-mip` might encounter `FileNotFoundError: HiGHS not found` even after `highspy` is installed. This suggests a solver discovery issue in the integrating library.
  Fix: Consult the documentation of the integrating modeling library. It may require setting an environment variable like `PMIP_HIGHS_LIBRARY` to explicitly point to the HiGHS shared library.
• deprecated The `addVar` method within the direct `highspy` modeling interface was noted for clashes and potential confusion with internal methods. A new method, `addVariable`, is being introduced for clearer high-level modeling.
  Fix: For high-level modeling, prefer `addVariable` and `addConstr` if available in your `highspy` version. For direct interaction with the solver's underlying data structures, use `addCol` and `addRow`.

Install

• pip install highspy Install with pip

Imports

• Highs

  import highspy
  h = highspy.Highs()

• HighsLp

  from highspy import HighsLp
  lp = HighsLp()

• ObjSense

  from highspy import ObjSense
  lp.sense_ = ObjSense.kMaximize

  Enums are accessed via the enum class (e.g., ObjSense) within the highspy module, not directly as module-level constants.

Quickstart

This quickstart demonstrates how to define and solve a simple Linear Programming (LP) problem using `highspy`. It constructs an `HighsLp` object, populates it with objective coefficients, variable bounds, and constraint matrix, then passes it to the HiGHS solver instance to find an optimal solution.

import highspy
import numpy as np

h = highspy.Highs()

# Define an LP problem (maximize 8x0 + 10x1 subject to constraints)
lp = highspy.HighsLp()
lp.num_col_ = 2
lp.num_row_ = 2
lp.sense_ = highspy.ObjSense.kMaximize
lp.col_cost_ = np.array([8, 10], dtype=np.double)
lp.col_lower_ = np.array([0, 0], dtype=np.double)
lp.col_upper_ = np.array([highspy.kHighsInf, highspy.kHighsInf], dtype=np.double)
lp.row_lower_ = np.array([-highspy.kHighsInf, -highspy.kHighsInf], dtype=np.double)
lp.row_upper_ = np.array([120, 210], dtype=np.double)

# Constraint matrix A (row-wise in this example for illustration)
# 0.3*x0 + 0.5*x1 <= 120
# 0.7*x0 + 0.5*x1 <= 210
lp.a_matrix_.start_ = np.array([0, 2, 4]) # Column starts (for column-wise storage, but example uses row-wise interpretation)
lp.a_matrix_.index_ = np.array([0, 1, 0, 1]) # Row indices for each element
lp.a_matrix_.value_ = np.array([0.3, 0.7, 0.5, 0.5], dtype=np.double)

h.passModel(lp)

# Solve the model
h.run()

# Extract and print solution
solution = h.getSolution()
info = h.getInfo()
model_status = h.getModelStatus()

print(f"Model status: {h.modelStatusToString(model_status)}")
if model_status == highspy.HighsModelStatus.kOptimal:
    col_value = list(solution.col_value)
    print(f"Optimal objective: {info.obj_val}")
    print(f"x0 = {col_value[0]}, x1 = {col_value[1]}")
else:
    print("No optimal solution found.")