All the hipster things in Neural Net in a single repo: hipster optimization algorithms, hispter regularizations, everything!
Note, things will be added over time, so not all the hipsterest things will be here immediately. Also don't use this for your production code: use this to study and learn new things in the realm of Neural Net, Deep Net, Deep Learning, whatever.
- Convolutional Net
- Feed Forward Net
- Recurrent Net
- LSTM Net
- GRU Net
- SGD
- Momentum SGD
- Nesterov Momentum
- Adagrad
- RMSprop
- Adam
- Cross Entropy
- Hinge Loss
- Squared Loss
- L1 Regression
- L2 Regression
- Dropout
- Your usual L1 and L2 regularization
- ReLU
- leaky ReLU
- sigmoid
- tanh
- BatchNorm
- Xavier weight initialization
- Max pooling
- Average pooling
- Install miniconda http://conda.pydata.org/miniconda.html
- Do
conda env create
- Enter the env
source activate hipsternet
- [Optional] To install Tensorflow:
chmod +x tensorflow.sh; ./tensorflow.sh
- Do things with the code if you want to
- To run the example:
python run_mnist.py {ff|cnn}
;cnn
for convnet model,ff
for the feed forward modelpython run_rnn.py {rnn|lstm|gru}
;rnn
for vanilla RNN model,lstm
for LSTM net model,gru
for GRU net model- Just close the terminal if you done (or
source deactivate
, not a fan though)
Do anything you want. I licensed this with Unlicense License http://unlicense.org, as I need to take a break of using WTFPL license.