DeepPy tries to combine state-of-the-art deep learning models with a Pythonic interface in an extensible framework.

• Pythonic programming interface based on NumPy's ndarray.
• Runs on CPU or Nvidia GPUs when available (thanks to CUDArray).
• Feedforward networks
  • Dropout layers.
  • Convnets layers: Convolution, pooling, local response normalization.
• Siamese Networks
• Training module
  • Stochastic gradient descent.
  • Interchangeable learning rules: Momentum, RMSProp.
  • Regularization: L2 weight decay.
• Dataset module
  • MNIST, CIFAR10

First, install CUDArray. Then install DeepPy with the standard

python setup.py install

• Dropout normalization of weights.
• Documentation!
• Support for regression problems in feed forward neural network.
• Other network types (autoencoders, stochastic neural networks, etc.).
• Interactive training method with visualization.

Thanks to the following projects for showing the way.

• scikit-learn
• Caffe
• Pylearn2