Comprehensive Guide to Resolving LAPACK/BLAS Resource Missing Issues in SciPy Installation on Windows

Problem Background and Error Analysis

When installing the SciPy scientific computing library on Windows operating systems, users frequently encounter the <span style="font-family: monospace;">numpy.distutils.system_info.NotFoundError: no lapack/blas resources found</span> error. The root cause of this issue lies in SciPy's core linear algebra modules depending on BLAS (Basic Linear Algebra Subprograms) and LAPACK (Linear Algebra PACKage) mathematical libraries, which are not included by default in Windows systems.

LAPACK/BLAS Dependency Analysis

BLAS and LAPACK are fundamental mathematical libraries in scientific computing, providing efficient implementations for core algorithms such as matrix operations, linear system solving, and eigenvalue computations. SciPy's linear algebra module (scipy.linalg) directly calls functions from these libraries to achieve high-performance numerical computing. While these libraries can be easily installed via package managers in Linux and macOS systems, additional configuration steps are required in Windows environments.

The search paths displayed in the error message indicate that the installation program looks for BLAS/LAPACK library files in multiple standard directories:

libraries mkl_rt not found in ['/QOpenSys/pkgs/lib', '/usr/local/lib', '/usr/lib64', '/usr/lib', '/usr/lib/']
libraries openblas not found in ['/QOpenSys/pkgs/lib', '/usr/local/lib', '/usr/lib64', '/usr/lib', '/usr/lib/']
libraries lapack not found in ['/QOpenSys/pkgs/lib', '/usr/local/lib', '/usr/lib64', '/usr/lib', '/usr/lib/']

Solution 1: Pre-compiled Binary Package Installation

For most Windows users, the simplest solution is to use pre-compiled binary packages. Christoph Gohlke's Python extension packages page (<span style="font-family: monospace;">http://www.lfd.uci.edu/~gohlke/pythonlibs/</span>) provides pre-compiled SciPy packages for different Python versions and system architectures.

Installation steps example:

# Download the wheel file corresponding to your version
# Example: scipy-1.8.0-cp39-cp39-win_amd64.whl

# Install the downloaded wheel file using pip
pip install scipy-1.8.0-cp39-cp39-win_amd64.whl

This method avoids complex compilation processes and directly provides complete installation packages containing optimized BLAS/LAPACK implementations. It's crucial to select wheel files that exactly match your Python version and system architecture.

Solution 2: Anaconda Scientific Computing Distribution

Anaconda is an integrated Python scientific computing distribution that comes with optimized mathematical libraries and dependency management. After installing Anaconda, SciPy and all its dependencies are automatically configured.

Advantages of Anaconda include:

• Automatic handling of all underlying dependencies
• Inclusion of Intel MKL optimized scientific computing libraries
• Provision of conda package manager for simplified environment management
• Support for virtual environment isolation

Installation command:

conda install scipy

While the basic version of Anaconda is free, certain advanced optimization features (such as GPU acceleration) require paid add-ons.

Solution 3: Manual Compilation and Performance Optimization

For users pursuing ultimate performance, manually compiling SciPy and linking optimized BLAS/LAPACK implementations is a worthwhile consideration. Intel Math Kernel Library (MKL) provides highly optimized mathematical function implementations that can significantly enhance scientific computing performance.

Intel MKL Integration

Intel MKL is a commercial mathematical library, but it can be legally used through Intel Parallel Studio's free trial or Intel Python distribution. Research shows that NumPy optimized with MKL can be up to 10 times faster than standard versions in large matrix operations.

Compilation configuration example:

# Set environment variables to point to MKL libraries
export BLAS="/path/to/mkl/libmkl_rt.so"
export LAPACK="/path/to/mkl/libmkl_rt.so"

# Compile and install SciPy
python setup.py build
python setup.py install

OpenBLAS Alternative

OpenBLAS is an open-source BLAS implementation with performance close to commercial libraries, and it's completely free. On Windows, it can be obtained through vcpkg or manual compilation:

# Install OpenBLAS using vcpkg
vcpkg install openblas

Development Environment Configuration Recommendations

For Python scientific computing development on Windows, the following toolchain configuration is recommended:

Compiler Toolset

Microsoft Visual C++ compiler is an essential compilation tool. For Python 3.4, the compiler corresponding to Visual Studio 2010 is required. Newer Python versions need corresponding versions of Visual Studio:

# Install Visual Studio C++ compiler
# Download URL: http://www.microsoft.com/en-us/download/details.aspx?id=44266

Build Tool Configuration

Properly configuring the numpy-site.cfg file can specify custom BLAS/LAPACK library paths:

# Example site.cfg configuration file
[openblas]
libraries = openblas
library_dirs = /path/to/openblas/lib
include_dirs = /path/to/openblas/include

Performance Comparison and Selection Recommendations

Different BLAS implementations have varying performance characteristics:

• Reference BLAS: Basic implementation, average performance
• OpenBLAS: Open-source optimization, excellent performance
• Intel MKL: Commercial-grade optimization, best performance
• ATLAS: Auto-tuning, moderate performance

Selection recommendations:

• Beginners: Recommend Anaconda or pre-compiled packages
• Regular users: Gohlke pre-compiled packages are sufficient
• Performance-sensitive users: Consider Intel MKL optimized versions
• Developers: Recommend configuring complete compilation environments

Troubleshooting and Verification

After installation, verify that SciPy is correctly installed and recognizes the BLAS/LAPACK backend using the following code:

import numpy as np
import scipy.linalg as la

# Test basic functionality
A = np.random.rand(100, 100)
# Eigenvalue computation
eigenvalues, eigenvectors = la.eig(A)
print("SciPy installation successful, eigenvalue computation normal")

# Check BLAS information
import scipy
print(f"SciPy version: {scipy.__version__}")
print(f"NumPy configuration: {np.__config__.show()}")

Conclusion and Outlook

The SciPy installation issues on Windows essentially represent management challenges for underlying dependencies in the scientific computing ecosystem. As the Python scientific computing ecosystem matures, increasingly more solutions are simplifying this process. In the future, with further development of package management tools and the proliferation of hardware acceleration technologies, the installation and usage of scientific computing libraries will become more convenient and efficient.

For specific application scenarios, users can also consider specialized scientific computing environments, such as MATLAB (with discounted pricing for students) or dedicated HPC solutions, which offer different balances between ease of use and performance.