README
¶
deepthought2
Neural network library for go. Version 2, work in progress.
Features:
- Intel MKL optimised matrix functions and DNN wrapper.
- Support for convolutional layers.
- Web based interface for plots and images.
- Example data models including MNist digit recognition.
To do:
- Edit layer setup from config screen
Dependencies:
- CPU Blas and DNN libraries https://software.intel.com/en-us/mkl
- GPU Blas and DNN libraries https://developer.nvidia.com/cublas https://developer.nvidia.com/cudnn
Package documentation: http://godoc.org/github.com/jnb666/deepthought2
Contact: John Banks jnb666@gmail.com
Directories
¶
| Path | Synopsis |
|---|---|
|
cifar-10 4 layer net based on https://blog.plon.io/tutorials/cifar-10-classification-using-keras-tutorial/ Simple 6 layer network based on https://www.kaggle.com/c/cifar-10/discussion/40237 Network in network: see https://arxiv.org/abs/1312.4400 Residual network: see https://arxiv.org/abs/1512.03385 Using simple architectures to outperform deeper and more complex architectures: see https://arxiv.org/pdf/1608.06037.pdf mnist conv net based on http://www.imm.dtu.dk/~abll/blog/simple_cnn/ mnist conv net based on http://neuralnetworksanddeeplearning.com/chap6.html mnist mlp exanple based on http://www.imm.dtu.dk/~abll/blog/simple_cnn/
|
cifar-10 4 layer net based on https://blog.plon.io/tutorials/cifar-10-classification-using-keras-tutorial/ Simple 6 layer network based on https://www.kaggle.com/c/cifar-10/discussion/40237 Network in network: see https://arxiv.org/abs/1312.4400 Residual network: see https://arxiv.org/abs/1512.03385 Using simple architectures to outperform deeper and more complex architectures: see https://arxiv.org/pdf/1608.06037.pdf mnist conv net based on http://www.imm.dtu.dk/~abll/blog/simple_cnn/ mnist conv net based on http://neuralnetworksanddeeplearning.com/chap6.html mnist mlp exanple based on http://www.imm.dtu.dk/~abll/blog/simple_cnn/ |
|
Package img contains routines for manipulating sets of images.
|
Package img contains routines for manipulating sets of images. |
|
Package nnet contains routines for constructing, training and testing neural networks.
|
Package nnet contains routines for constructing, training and testing neural networks. |
|
Package num contains numeric Array processing routines such as optimised matix multiplication.
|
Package num contains numeric Array processing routines such as optimised matix multiplication. |
|
cuda
Package cuda contains wrapper functions for Cuda api
|
Package cuda contains wrapper functions for Cuda api |
|
mkl
Package mkl wraps wraps the Intel MKL DNN functions
|
Package mkl wraps wraps the Intel MKL DNN functions |
|
Package web has a web based interface we network training and visualisation.
|
Package web has a web based interface we network training and visualisation. |
Click to show internal directories.
Click to hide internal directories.