개인적으로 Torch를 주력으로 사용하려합니다. 필요할 때마다 빠르게 검색하기 위해 모아놓은 것들을 wiki 처럼 공유합니다.

Official Homepage

• Installation is easy.

Tutorials

Very useful

15 min Lua

Torch Video Tutorials

Bayarea DL summmer

slide

ref when needed

demos

DOCS

tutorial

Useful Libraries

Torchnet

• torchnet is a framework for torch which provides a set of abstractions aiming at encouraging code re-use as well as encouraging modular programming.
  At the moment, torchnet provides four set of important classes:
  Dataset: handling and pre-processing data in various ways.
  Engine: training/testing machine learning algorithm.
  Meter: meter performance or any other quantity.
  Log: output performance or any other string to file / disk in a consistent manner.
  For an overview of the torchnet framework, please also refer to this paper.

Facebook : lua library

• Includes debugger (very useful)

Deepmind : Cephes Mathematical Funcions Library

• Provides and wraps the mathematical functions from the Cephes mathematical library, developed by Stephen L. Moshier. This C library provides a lot of mathematical functions. It is used, among many other places, at the heart of SciPy.

Twitter Autograd

• compute gradient automatically
• similar to tensorflow, theano auto differentiation.
• Useful to design very wierd architectures.
• ref: Bayarea DL summmer

grad = require 'autograd'
grad.optimize(true) -- global
local df = grad(f, { optimize = true }) -- for this function only
local grads = df(params)

conda

OptNet - reducing memory usage in torch neural networks

• reduce memory

Display

• Web broswer
• Useful with SSH
• Almost only GUI with torch.
• socket based. Listening at certain port
• Developed from gfx.js : a browser-based graphics server
• Depracated. Can use terminal

From Torch.ch

• torch
• nn
• optim
• gnuplot
• paths
• image
• trepl
• cwrap
• qtlua
• awsome-torch

Neural Network Package

This package provides an easy and modular way to build and train simple or complex neural networks using Torch:

• Modules are the bricks used to build neural networks. Each are themselves neural networks, but can be combined with other networks using containers to create complex neural networks:
  • [Module]
    (https://github.com/torch/nn/blob/master/doc/module.md#nn.Module): abstract class inherited by all modules;
  • Containers: container classes like Sequential, Parallel and Concat;
  • Transfer functions: non-linear functions like Tanh and Sigmoid;
  • Simple layers: like Linear, Mean, Max and Reshape;
  • Table layers: layers for manipulating tables like SplitTable, ConcatTable and JoinTable;
  • Convolution layers: Temporal, Spatial and Volumetric convolutions;
• Criterions compute a gradient according to a given loss function given an input and a target:
  • Criterions: a list of all criterions, including Criterion, the abstract class;
  • MSECriterion: the Mean Squared Error criterion used for regression;
  • ClassNLLCriterion: the Negative Log Likelihood criterion used for classification;
• Additional documentation:
  • Overview of the package essentials including modules, containers and training;
  • Training: how to train a neural network using StochasticGradient;
  • Testing: how to test your modules.
  • Experimental Modules: a package containing experimental modules and criteria.