yadll

Yet Another Deep Learning Lab. Ultra light Deep Learning framework based on Theano


Keywords
deep-learning, theano
License
MIT
Install
pip install yadll==0.0.1

Documentation

https://coveralls.io/repos/github/pchavanne/yadll/badge.svg?branch=master https://readthedocs.org/projects/yadll/badge/

Yadll

Yet another deep learning lab.

This is an ultra light deep learning framework written in Python and based on Theano. It allows you to very quickly start building Deep Learning models. It was originally the code, notes and references I gathered when following the Theano's Deep Learning Tutorials tutorial then I used Lasagne, keras or blocks and restructured this code based on it.

If you are looking for a light deep learning API I would recommend using Lasagne or keras in stead of yadll, both are mature, well documented and contributed projects.

Read the documentation at Read the doc

Its main features are:

  • Layers:
    • Input Layer
    • Dropout Layer
    • Pool Layer
    • Conv Layer:
      • ConvPool Layer
    • Dense Layer:
      • Logistic Regression
      • Dropconnect
      • Unsupervised Layer:
        • Autoencoder (denoising autoencoder)
        • Restricted Boltzmann Machine
    • RNN
    • LSTM
  • Optimisation:
    • Sgd
    • Momentum
    • Nesterov momentum
    • Adagrad
    • Adadelta
    • Rmsprop
    • Hessian Free
  • Hyperparameters grid search

Installation

git clone git@github.com:pchavanne/yadll.git
cd yadll
pip install -e .

Example

Different networks tested on mnist:

  • Logisitic Regression
  • Multi Layer Perceptron
  • MLP with dropout
  • MLP with dropconnect
  • Conv Pool
  • LeNet-5
  • Autoencoder
  • Denoising Autoencoder
  • Gaussian Denoising Autoencoder
  • Contractive Denoising Autoencoder
  • Stacked Denoising Autoencoder
  • Restricted Boltzmann Machine
  • Deep Belief Network
  • Recurent Neural Networks
  • Long Short-Term Memory

get the list of available networks:

python mnist_dl.py --network_list

trainning a model for example lenet5:

python mnist_dl.py lenet5

grid search on the hyperparameters:

python hp_grid_search.py