search

shogun-toolbox / shogun

Shōgun
http://shogun-toolbox.org
BSD 3-Clause "New" or "Revised" License
3.03k stars 1.04k forks source link

Populate and extend dimensionality reduction notebook #2674

Open karlnapf opened 9 years ago

karlnapf commented 9 years ago

also related to #2673

srgnuclear commented 9 years ago

List of dimensionality reduction algorithms - PCA SPE(Stochastic proximity embedding) Diffusion Maps Kernel Locally Linear Embedding Kernel Local Tangent Space Alignment Linear Local Tangent Space Alignment Neighborhood Preserving embedding Locality Preserving Projections Locally Linear Embedding (LLE) Hessian Locally Linear Embedding (HLLE) Local Tangent Space Alignment (LTSA) Kernel PCA (kPCA) Multidimensional Scaling (MDS, with possible landmark approximation) Isomap (using Fibonacci Heap Dijkstra for shortest paths) Laplacian Eigenmaps.

Do you think their could be more? I made search for dimensionality reduction in the repo and found these algorithms.

srgnuclear commented 9 years ago

By reproduce common examples do you mean various examples coded in python in scikit learn to be recoded in cpp and put in shogun or anything else?

karlnapf commented 9 years ago

Hi, thanks for the list! I mean to reproduce standard examples in the notebook -- using Shogun No cpp involved

sudk1896 commented 7 years ago

@karlnapf : I recreated one of the listed tutorials here using SHOGUN (in python). I used the USPS dataset and created an analog of the scikit tutorial. More specifically on Hessian LLE given here.

Thing is how do I know if this is correct/accurate. I can't find the SHOGUN analog of recontruction_error in the Hessian LLE module (Check the scikit page). Let me know If you'd like to take a look at my notebook. I will create more which don't have any examples. I had to slice the USPS dataset and make it smaller to run(Hope that is okay). The scikit example btw uses a smaller dataset as well. The reconstruction looks like this, I changed the dimensions from (256, 9000) -> (256, 2048) to make it feasible to run. The notebook is here

figure_1

karlnapf commented 7 years ago

Nice in principle. Can you send a PR then I can give comments. Please always paste a link of the notebook with outputs in the PR. The PR itself should be without outputs. Looking forward to reviewing it

@lisitsyn any ideas on the reconstruction error? If we dont have it, would be good to add that to shogun, it is easy