/structure-random-projection

structure-random-projection

Primary LanguageJupyter Notebook

Tensor Random Projection

Introduction

In this project, we implement the Tensor Random Projection based on the Khatri-Rao product and extend it to matrix sketching. We run the experiment with simiulated data and the MNIST data.

Files

  1. util.py: Helper functions for data generation, variance reduction, and others.
  2. simulation.py: Function for creating simulation for both random projection and sketching.
  3. nips_simulation.py: Run experiments for norm preservation and sketching with simulated data on server.
  4. minst.py: Evaluate the result for inner product preservation with MNIST data on server.
  5. nips_simulation.ipynb: Evaluate the norm preservation and sketching for simulated data, and inner product preservation for MNIST data (Replicate Table 1,2, Figure 2-6)
  6. dist_simulation.ipynb: Evaluate the distance preservation for both simulated and MNIST data. (Replicate Figure 1)

Todo

  1. Applied to Kernel regression with Kronecker product