misso 0.1

MISSO

Installation

• Using pip

pip install misso

Note: In our benchmarks, multi-core version was always faster than the GPU accelerated version. So, we highly recommend installing just the CPU version and using multi-core computation.

• Installing from source

Usage

from misso import MISSO

For a more detailed usage, check out the Tutorials folder.

Benchmarks

** Benchmarks were run on a machine with the following configuration

CPU:       6 core Intel Core i7-8750H (-MT-MCP-) [12 core with Hyperthreading]
           arch: Skylake rev.10 cache: 9216 KB
           flags: (lm nx sse sse2 sse3 sse4_1 sse4_2 ssse3 vmx) bmips: 26399
           clock speeds: max: 4100 MHz 1: 2479 MHz 2: 3013 MHz 3: 3211 MHz
           4: 3098 MHz 5: 3362 MHz 6: 3769 MHz 7: 3082 MHz 8: 3290 MHz
           9: 3090 MHz 10: 3141 MHz 11: 3055 MHz 12: 3650 MHz
Graphics:  Card-1: Intel Device 3e9b bus-ID: 00:02.0
           Card-2: NVIDIA Device 1f10 bus-ID: 01:00.0
           Display Server: x11 (X.Org 1.19.6 )
           drivers: modesetting,nvidia (unloaded: fbdev,vesa,nouveau)
           Resolution: 3840x1600@59.99hz
           OpenGL: renderer: GeForce RTX 2070 with Max-Q Design/PCIe/SSE2
           version: 4.6.0 NVIDIA 440.100 Direct Render: Yes

License

TODO

• Try gradient-based solvers
  • Conjugate-gradient descent
• Multi-processing for lsmi computation
  • Reduce interprocess overhead
  • Try other methods to parallelize the code
• Benchmarks
  • Multiprocessing
  • GPU benchmarks
  • Solver Benchmarks
  • Run benchmarks on multiple machines and put in benchmark reports
• Detailed comparison with graphical Lasso (Tutorials)
  • Toy Example
  • Time Series: Stationary & Dynamic link
  • Comparison of MISSO and GLASSO on indirect coupling link
• GPU Acceleration
  • Use Cupy for solving
  • Reduce GPU overhead
  • Verify correctness (Still an issue)
  • Try torch for GPU acceleration
• tqdm for Notebook and Script
• Pandas DataFrame support
• Packaging
  • pip package
  • Travis CI
• Readme