This is a package to compare compuatational time of different packages in Python for solving eigenvalue problems for a sparse matrix. The sparse matrix is generated
using scipy.sparse.random and converted to its dense form using scipy.sparse.csr_matrix.todense function. This matrix is used to solve the
eigenvalue problems using the following packages
numpy.linalg.eighscipy.linalg.eighscipy.sparse.linalg.eigshtensorflow.linalg.eigh
Run the following snippet inside the test_eigprob directory from your terminal:
python run_time_eigprob.py
The timing of matrices of different sizes on 10 CPU cores and 1 GPU on Princeton's Tiger Cluster are tabulated below:
| Function | Time taken (in seconds) |
|---|---|
| Matrix size (10 x 10) | |
| numpy.linalg.eigh | 0.0000 |
| scipy.linalg.eigh | 0.0001 |
| scipy.sparse.linalg.eigsh | 0.0004 |
| tensorflow.linalg.eigh | 0.0004 |
| Matrix size (31 x 31) | |
| numpy.linalg.eigh | 0.0000 |
| scipy.linalg.eigh | 0.0001 |
| scipy.sparse.linalg.eigsh | 0.0007 |
| tensorflow.linalg.eigh | 0.0008 |
| Matrix size (100 x 100) | |
| numpy.linalg.eigh | 0.0005 |
| scipy.linalg.eigh | 0.0007 |
| scipy.sparse.linalg.eigsh | 0.0009 |
| tensorflow.linalg.eigh | 0.0042 |
| Matrix size (316 x 316) | |
| numpy.linalg.eigh | 0.0052 |
| scipy.linalg.eigh | 0.0087 |
| scipy.sparse.linalg.eigsh | 0.0134 |
| tensorflow.linalg.eigh | 0.0209 |
| Matrix size (1000 x 1000) | |
| numpy.linalg.eigh | 0.0540 |
| scipy.linalg.eigh | 0.1570 |
| scipy.sparse.linalg.eigsh | 0.0275 |
| tensorflow.linalg.eigh | 0.0859 |
| Matrix size (3162 x 3162) | |
| numpy.linalg.eigh | 1.0206 |
| scipy.linalg.eigh | 2.1212 |
| scipy.sparse.linalg.eigsh | 0.7109 |
| tensorflow.linalg.eigh | 0.4971 |