Related projects

Frédéric Bastien edited this page Sep 11, 2017 · 61 revisions

This list is far from complete. Please feel free to add to this list!

Parallelism

  • Synkhronos Extension to Theano for multi-GPU data parallelism
  • Theano-MPI Theano-MPI a distributed framework for training models built in Theano based on data-parallelism.
  • Platoon Multi-GPU mini-framework for Theano, single node.
  • Elephas Distributed Deep Learning with Keras & Spark.

Cloud image, VM or container

GUI based interface

Tutorials and sample code

Libraries built on Theano

  • Lasagne: a lightweight library to build and train neural networks in Theano, with a focus on feed-forward neural networks but there are extensions for RNNs including LSTMs.
  • Keras: a minimalist, highly modular neural network library in the spirit of Torch.
  • rllab: rllab is a framework for developing and evaluating reinforcement learning algorithms, fully compatible with OpenAI Gym.
  • Blocks: a framework that helps you build neural network models on top of Theano. Includes RNNs.
  • PyMC 3.0: Bayesian statistical modeling and model fitting which focuses on advanced Markov chain Monte Carlo fitting algorithms.
  • DeepMedic: or brain lesion segmentation based on a multi-scale 3D Deep Convolutional Neural Network
  • sklearn-theano: A sklearn interface for preprocessing using learnt models that use Theano.
  • Downhill: Provides algorithms for minimizing scalar loss for Theano functions.
  • Theano_lgpl: Theano extension that use lgpl code. See that page for a list what is add.
  • Mariana: A machine learning framework that aims at greatly simplifying the writing and handling of neural networks (Especially deep ones).
  • Passage: A little library for text analysis with RNNs.
  • PyAutoDiff: Automatic differentiation for NumPy code.
  • theano-rnn theano-rnn: Recurrent neural network implementation. (2 differents implementations)
  • Morb: a modular RBM implementation in Theano
  • MonteTheano: Sampling for directed graphical models, more probability distributions than in Theano itself.
  • Ape(status alpha): Theano/MPI bridge which will also try to statically schedule the operation onto heterogeneous hardware.
  • Crino: a neural-network library based on Theano
  • Theanet: Theano based Convolutional Neural Network for image classification with Elastic Distortion & Noising of inputs, Dropout, Maxnorm Regularization, Softmax, Mixture of Gaussians outputs etc.
  • Merlin: Merlin is a toolkit for building Deep Neural Network models for statistical parametric speech synthesis.
  • Kaldi+PDNN Kaldi+PDNN builds state-of-the-art DNN acoustic models using the open-source Kaldi and PDNN toolkits
  • theano-lstm
  • Plato: A simplified API and deep learning library built on top of Theano. Includes MLPs, RBMs, Deep Belief Nets, LSTMs, Variational Autoencoders, Difference Target Prop, etc. Tutorial
  • OpenDeep a Torch-like modular framework built on Theano with included models for CNNs, RBMs, RNNs, GSNs, LSTMs and more coming. Current status of the software is alpha.
  • nnet-ts Neural network architecture for time series forecasting.
  • pyfolio A Python library for performance and risk analysis of financial portfolios (use PyMC3 that use Theano)
  • tmetrics A collections of metrics and loss functions written in Theano
  • pymanopt Manifold optimization using Theano for gradient (and Hessian) calculations, and code ported from manopt for the optimization.
  • ELEKTRONN A highly configurable toolkit with focus on high throughput analysis of large scale 2d and 3d images with convolutional neural networks e.g. in connectomics applications.
  • deepy: a deep learning framework for designing models with complex architectures.
  • Theano-Lights: a research framework based on Theano providing implementation of several recent Deep learning models and a convenient training and test functionality
  • yann: a research framework and a toolbox on convolutional neural networks based on Theano with helpful tutorials for students to learn from.

Models (not all models, only a few. Frameworks based on Theano have more)

Applications

Toolkits / ops

  • Artificial Dataset Generation: this dataset generation can be used to do emperical measurements of Machine Learning algorithms.
  • Hessian Free: Code to implement Hessian-Free model
  • Janus: Janus is a tool that allows NumPy and Theano to be used simultaneously with no additional code.
  • Kaldi wrapper They require boost::python

Some specific function requested on the mailing list

Deprecated software

  • GroundHog: An RNN framework.
  • Pylearn2: A machine learning library.
  • DeepANN: Deep ANN implementation.
  • TheanoConv3d2d: (Now integrated in Theano) A GPU-friendly 3D convolution implementation based on Conv2D.

Github Search

Didn't find what you were looking for? Try a github search: