Parallel t-SNE implementation with Python and Torch wrappers.

Overview

Multicore t-SNE Build Status

This is a multicore modification of Barnes-Hut t-SNE by L. Van der Maaten with python and Torch CFFI-based wrappers. This code also works faster than sklearn.TSNE on 1 core.

What to expect

Barnes-Hut t-SNE is done in two steps.

  • First step: an efficient data structure for nearest neighbours search is built and used to compute probabilities. This can be done in parallel for each point in the dataset, this is why we can expect a good speed-up by using more cores.

  • Second step: the embedding is optimized using gradient descent. This part is essentially consecutive so we can only optimize within iteration. In fact some parts can be parallelized effectively, but not all of them a parallelized for now. That is why second step speed-up will not be that significant as first step sepeed-up but there is still room for improvement.

So when can you benefit from parallelization? It is almost true, that the second step computation time is constant of D and depends mostly on N. The first part's time depends on D a lot, so for small D time(Step 1) << time(Step 2), for large D time(Step 1) >> time(Step 2). As we are only good at parallelizing step 1 we will benefit most when D is large enough (MNIST's D = 784 is large, D = 10 even for N=1000000 is not so much). I wrote multicore modification originally for Springleaf competition, where my data table was about 300000 x 3000 and only several days left till the end of the competition so any speed-up was handy.

Benchmark

1 core

Interestingly, that this code beats other implementations. We compare to sklearn (Barnes-Hut of course), L. Van der Maaten's bhtsne, py_bh_tsne repo (cython wrapper for bhtsne with QuadTree). perplexity = 30, theta=0.5 for every run. In fact py_bh_tsne repo works at the same speed as this code when using more optimization flags for compiler.

This is a benchmark for 70000x784 MNIST data:

Method Step 1 (sec) Step 2 (sec)
MulticoreTSNE(n_jobs=1) 912 350
bhtsne 4257 1233
py_bh_tsne 1232 367
sklearn(0.18) ~5400 ~20920

I did my best to find what is wrong with sklearn numbers, but it is the best benchmark I could do (you can find test script in python/tests folder).

Multicore

This table shows a relative to 1 core speed-up when using n cores.

n_jobs Step 1 Step 2
1 1x 1x
2 1.54x 1.05x
4 2.6x 1.2x
8 5.6x 1.65x

How to use

Python and torch wrappers are available.

Python

Install

Directly from pypi

pip install MulticoreTSNE

From source

Make sure cmake is installed on your system, and you will also need a sensible C++ compiler, such as gcc or llvm-clang. On macOS, you can get both via homebrew.

To install the package, please do:

git clone https://github.com/DmitryUlyanov/Multicore-TSNE.git
cd Multicore-TSNE/
pip install .

Tested with both Python 2.7 and 3.6 (conda) and Ubuntu 14.04.

Run

You can use it as a near drop-in replacement for sklearn.manifold.TSNE.

from MulticoreTSNE import MulticoreTSNE as TSNE

tsne = TSNE(n_jobs=4)
Y = tsne.fit_transform(X)

Please refer to sklearn TSNE manual for parameters explanation.

This implementation n_components=2, which is the most common case (use Barnes-Hut t-SNE or sklearn otherwise). Also note that some parameters are there just for the sake of compatibility with sklearn and are otherwise ignored. See MulticoreTSNE class docstring for more info.

MNIST example

from sklearn.datasets import load_digits
from MulticoreTSNE import MulticoreTSNE as TSNE
from matplotlib import pyplot as plt

digits = load_digits()
embeddings = TSNE(n_jobs=4).fit_transform(digits.data)
vis_x = embeddings[:, 0]
vis_y = embeddings[:, 1]
plt.scatter(vis_x, vis_y, c=digits.target, cmap=plt.cm.get_cmap("jet", 10), marker='.')
plt.colorbar(ticks=range(10))
plt.clim(-0.5, 9.5)
plt.show()

Test

You can test it on MNIST dataset with the following command:

python MulticoreTSNE/examples/test.py <n_jobs>

Note on jupyter use

To make the computation log visible in jupyter please install wurlitzer (pip install wurlitzer) and execute this line in any cell beforehand:

%load_ext wurlitzer

Memory leakages are possible if you interrupt the process. Should be OK if you let it run until the end.

Torch

To install execute the following command from repository folder:

luarocks make torch/tsne-1.0-0.rockspec

or

luarocks install https://raw.githubusercontent.com/DmitryUlyanov/Multicore-TSNE/master/torch/tsne-1.0-0.rockspec

You can run t-SNE like that:

tsne = require 'tsne'

Y = tsne(X, n_components, perplexity, n_iter, angle, n_jobs)

torch.DoubleTensor type only supported for now.

License

Inherited from original repo's license.

Future work

  • Allow other types than double
  • Improve step 2 performance (possible)

Citation

Please cite this repository if it was useful for your research:

@misc{Ulyanov2016,
  author = {Ulyanov, Dmitry},
  title = {Multicore-TSNE},
  year = {2016},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/DmitryUlyanov/Multicore-TSNE}},
}

Of course, do not forget to cite L. Van der Maaten's paper

Owner
Dmitry Ulyanov
Co-Founder at in3D, Phd @ Skoltech
Dmitry Ulyanov
Smarthome Dashboard with Grafana & InfluxDB

Smarthome Dashboard with Grafana & InfluxDB This is a complete overhaul of my Raspberry Dashboard done with Flask. I switched from sqlite to InfluxDB

6 Oct 20, 2022
Application for viewing pokemon regional variants.

Pokemon Regional Variants Application Application for viewing pokemon regional variants. Run The Source Code Download Python https://www.python.org/do

Michael J Bailey 4 Oct 08, 2021
Python ts2vg package provides high-performance algorithm implementations to build visibility graphs from time series data.

ts2vg: Time series to visibility graphs The Python ts2vg package provides high-performance algorithm implementations to build visibility graphs from t

Carlos Bergillos 26 Dec 17, 2022
A tool to plot and execute Rossmos's Formula, that helps to catch serial criminals using mathematics

Rossmo Plotter A tool to plot and execute Rossmos's Formula using python, that helps to catch serial criminals using mathematics Author: Amlan Saha Ku

Amlan Saha Kundu 3 Aug 29, 2022
📊 Charts with pure python

A zero-dependency python package that prints basic charts to a Jupyter output Charts supported: Bar graphs Scatter plots Histograms 🍑 📊 👏 Examples

Max Humber 54 Oct 04, 2022
Tandem Mass Spectrum Prediction with Graph Transformers

MassFormer This is the original implementation of MassFormer, a graph transformer for small molecule MS/MS prediction. Check out the preprint on arxiv

Röst Lab 13 Oct 27, 2022
Realtime Viewer Mandelbrot set with Python and Taichi (cpu, opengl, cuda, vulkan, metal)

Mandelbrot-set-Realtime-Viewer- Realtime Viewer Mandelbrot set with Python and Taichi (cpu, opengl, cuda, vulkan, metal) Control: "WASD" - movement, "

22 Oct 31, 2022
Data visualization using matplotlib

Data visualization using matplotlib project instructions Top 5 Most Common Coffee Origins In this visualization I used data from Ankur Chavda on Kaggl

13 Oct 27, 2021
在原神中使用围栏绘图

yuanshen_draw 在原神中使用围栏绘图 文件说明 toLines.py 将一张图片转换为对应的线条集合,视频可以按帧转换。 draw.py 在原神家园里绘制一张线条图。 draw_video.py 在原神家园里绘制视频(自动按帧摆放,截图(win)并回收) cat_to_video.py

14 Oct 08, 2022
This tool is designed to help administrators get an overview of their Active Directory structure.

This tool is designed to help administrators get an overview of their Active Directory structure. In the group view you can see all elements of an AD (OU, USER, GROUPS, COMPUTERS etc.). In the user v

deexno 2 Oct 30, 2022
Python package for the analysis and visualisation of finite-difference fields.

discretisedfield Marijan Beg1,2, Martin Lang2, Samuel Holt3, Ryan A. Pepper4, Hans Fangohr2,5,6 1 Department of Earth Science and Engineering, Imperia

ubermag 12 Dec 14, 2022
NumPy and Pandas interface to Big Data

Blaze translates a subset of modified NumPy and Pandas-like syntax to databases and other computing systems. Blaze allows Python users a familiar inte

Blaze 3.1k Jan 01, 2023
Using SQLite within Python to create database and analyze Starcraft 2 units data (Pandas also used)

SQLite python Starcraft 2 English This project shows the usage of SQLite with python. To create, modify and communicate with the SQLite database from

1 Dec 30, 2021
Graphical visualizer for spectralyze by Lauchmelder23

spectralyze visualizer Graphical visualizer for spectralyze by Lauchmelder23 Install Install matplotlib and ffmpeg. Put ffmpeg.exe in same folder as v

Matthew 1 Dec 21, 2021
Make sankey, alluvial and sankey bump plots in ggplot

The goal of ggsankey is to make beautiful sankey, alluvial and sankey bump plots in ggplot2

David Sjoberg 156 Jan 03, 2023
Shaded 😎 quantile plots

shadyquant 😎 This python package allows you to quantile and plot lines where you have multiple samples, typically for visualizing uncertainty. Your d

Mehrad Ansari 13 Sep 29, 2022
Piglet-shaders - PoC of custom shaders for Piglet

Piglet custom shader PoC This is a PoC for compiling Piglet fragment shaders usi

6 Mar 10, 2022
NorthPitch is a python soccer plotting library that sits on top of Matplotlib

NorthPitch is a python soccer plotting library that sits on top of Matplotlib.

Devin Pleuler 30 Feb 22, 2022
Simple and fast histogramming in Python accelerated with OpenMP.

pygram11 Simple and fast histogramming in Python accelerated with OpenMP with help from pybind11. pygram11 provides functions for very fast histogram

Doug Davis 28 Dec 14, 2022
Mathematical learnings with Lean, for those of us who wish we knew more of both!

Lean for the Inept Mathematician This repository contains source files for a number of articles or posts aimed at explaining bite-sized mathematical c

Julian Berman 8 Feb 14, 2022