Skip to content
This repository has been archived by the owner on Jan 13, 2022. It is now read-only.
/ iTorch Public archive

IPython kernel for Torch with visualization and plotting

License

Notifications You must be signed in to change notification settings

facebookarchive/iTorch

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

iTorch

iTorch is an IPython Kernel for Torch, with plotting (using Bokeh.js plots) and visualization of images, video and audio

Features

iTorch in notebook mode works like any other IPython notebook.
It provides useful inline auto-complete. Whenever you need auto-complete, use the TAB key.
screenshot

It also provides inline help using the ? symbol. For example, ?torch.cmul screenshot

In addition, we introduce visualization functions for images, video, audio, html and plots.

itorch.image(img) - You can pass in a 3-D tensor (for a single image), or a table of 3D tensors (image collages).

  itorch.image({image.lena(), image.lena(), image.lena()})

screenshot

itorch.audio(path) - You can pass in a filename of an audio file. Formats supported are mp3, ogg, aac

itorch.audio('example.mp3')

screenshot

itorch.video(path) - You can pass in a filename of a video file. Formats supported are mp4, ogv, mpeg

itorch.video('example.mp4')

screenshot

[window-id] = itorch.html(htmlstring, [window-id]) - Raw HTML string that is passed is rendered. A window handle is returned, that can be reused to replace the HTML with something else.

itorch.html('<p><b>Hi there!</b> this is arbitrary HTML</p>')
window_id = itorch.html('<p>This text will be replaced in 2 seconds</p>')
os.execute('sleep 2')
itorch.html('<p>magic!</p>', window_id)

screenshot

###Plotting iTorch can plot to screen in notebook mode, or save the plot to disk as a html file.

A Plot object is introduced, that can plot different kinds of plots such as scatter, line, segment, quiver plots.

Plot = require 'itorch.Plot'

The plotting can be extended to more kinds of plots, as it uses Bokeh.js as its backend.

x1 = torch.randn(40):mul(100)
y1 = torch.randn(40):mul(100)
x2 = torch.randn(40):mul(100)
y2 = torch.randn(40):mul(100)
x3 = torch.randn(40):mul(200)
y3 = torch.randn(40):mul(200)


-- scatter plots
plot = Plot():circle(x1, y1, 'red', 'hi'):circle(x2, y2, 'blue', 'bye'):draw()
plot:circle(x3,y3,'green', 'yolo'):redraw()
plot:title('Scatter Plot Demo'):redraw()
plot:xaxis('length'):yaxis('width'):redraw()
plot:legend(true)
plot:redraw()
-- print(plot:toHTML())
plot:save('out.html')

screenshot

-- line plots
plot = Plot():line(x1, y1,'red','example'):legend(true):title('Line Plot Demo'):draw()
-- segment plots
plot = Plot():segment(x1, y1, x1+10,y1+10, 'red','demo'):title('Segment Plot Demo'):draw()
-- quiver plots
U = torch.randn(3,3):mul(100)
V = torch.randn(3,3):mul(100)
plot = Plot():quiver(U,V,'red',''):title('Quiver Plot Demo'):draw()

screenshot

-- quads/rectangles
x1=torch.randn(10)
y1=torch.randn(10)
plot = Plot():quad(x1,y1,x1+1,y1+1,'red',''):draw()
-- histogram
plot = Plot():histogram(torch.randn(10000)):draw()

screenshot

Hover Tool in Plotting

local t = torch.Tensor
local y = t(10)
local x = t(y:size()):zero()
local labels = {}
for i = 1, 10 do
    y[i] = i
	labels[i] = tostring(i)
end

itorch.Plot()
  :circle(x, y, 'red', nil, {foo=labels})
  :hover_tool({{'xy', '@x @y'}, {'foo', '@foo'}})
  :draw()

Text method to plotting

local t = torch.Tensor
local y = t(10)
local x = t(y:size()):zero()
local labels = {}
for i = 1, 10 do
    y[i] = i
	labels[i] = tostring(i)
end

itorch.Plot():gscatter(x, y)
  :text(x, y, labels, y, 'black')
  :triangle(x, y, 'blue')
  :draw()

Group Scatter plot

Run the following in itorch to produce plots:

x = torch.randn(200); y = torch.randn(200); x:narrow(1, 1, 100):add(2);
labels = torch.LongTensor(200):zero(); labels:add(1); labels:narrow(1, 1, 100):add(1)
itorch.Plot():gscatter(x, y):title('Scatter plot without labels'):draw()
itorch.Plot():gscatter(x, y, labels):title('Scatter plot with labels and legend #1'):legend(true):draw()
itorch.Plot():gscatter(x, y, labels, true):title('Scatter plot with labels and legend #2'):draw()
itorch.Plot():gscatter(x, y, labels, false):title('Scatter plot with labels and no legend'):draw()

Requirements

iTorch requires or works with

  • Mac OS X or Linux (tested in Ubuntu 14.04 and Arch Linux)
  • Torch-7
  • IPython version 2.2 or above (you can check your version of ipython using ipython --version)
  • ZeroMQ
# OSX
brew install zeromq
brew install openssl
luarocks install luacrypto OPENSSL_DIR=/usr/local/opt/openssl/

# Ubuntu
sudo apt-get install libzmq3-dev libssl-dev python-zmq

# Ubuntu 16
luarocks install lzmq

Installing iTorch

git clone https://github.com/facebook/iTorch.git
cd iTorch
luarocks make 

If you have to use sudo for some reason (if you globally installed torch on Linux for example), use these commands:

sudo env "PATH=$PATH" luarocks make
sudo chown -R $USER $(dirname $(ipython locate profile))

Docker Images

Ubuntu 14.04 + iTorch notebook + Miniconda: see docker hub repo.

docker pull dhunter/itorch-notebook

How iTorch works

Start iTorch at command-line using the following command:

itorch notebook  # notebook mode
OR  
itorch  # console mode
OR  
itorch qtconsole  # Qt mode

In notebook mode, you can use the "New" button to create a new notebook.

Examples

Demo iTorch notebook: http://nbviewer.ipython.org/github/facebook/iTorch/blob/master/iTorch_Demo.ipynb

Join the Torch community

See the CONTRIBUTING file for how to help out.

License

iTorch is BSD-licensed. We also provide an additional patent grant.

About

IPython kernel for Torch with visualization and plotting

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published