deepnet
Implementations of CNNs, RNNs and cool new techniques in deep learning
Note: deepnet is a work in progress and things will be added gradually. It is not intended for production, use it to learn and study implementations of latest and greatest in deep learning.
What does it have?
Network Architecture
- Convolutional net
- Feed forward net
- Recurrent net (LSTM/GRU coming soon)
Optimization Algorithms
- SGD
- SGD with momentum
- Nesterov Accelerated Gradient
- Adagrad
- RMSprop
- Adam
Regularization
- Dropout
- L1 and L2 Regularization
Cool Techniques
- BatchNorm
- Xavier Weight Initialization
Nonlinearities
- ReLU
- Sigmoid
- tanh
Usage
virtualenv .env
; create a virtual environmentsource .env/bin/activate
; activate the virtual environmentpip install -r requirements.txt
; Install dependenciespython run_cnn.py {mnist|cifar10}
; mnist for shallow cnn and cifar10 for deep cnn