Tyill/skynet

neural net with blackjack and hookers


repo name	Tyill/skynet
repo link	https://github.com/Tyill/skynet
homepage
language	C++
size (curr.)	90124 kB
stars (curr.)	53
created	2018-08-13
license	MIT License

SkyNet is a light deep learning library.

`Linux/Windows`	`License`

ResNet cpp-example for Win

Compare with Tensorflow, inference ResNet50. PC: i5-2400, GF1050, Win7, MSVC12.

	CPU: time/img, ms	GPU: time/img, ms	CPU: RAM, Mb	GPU: RAM, Mb
Skynet	195	15	600	800
Tensorflow	250	25	400	1400

Features

the library is written from scratch in C++ (only STL + OpenBLAS for calculation), C-interface
win / linux;
network structure is set in JSON;
base layers: fully connected, convolutional, pooling. Additional: resize, crop..;
basic chips: batchNorm, dropout, weight optimizers - adam, adagrad..;
for calculation on the CPU, OpenBLAS is used, for the video card - CUDA / cuDNN;
interfaces for C++, C# and Python.

Python example


# create net
net = snNet.Net()
net.addNode('In', snOperator.Input(), 'C1') \
   .addNode('C1', snOperator.Convolution(15), 'C2') \
   .addNode('C2', snOperator.Convolution(25), 'P1') \
   .addNode('P1', snOperator.Pooling(), 'F1') \
   .addNode('F1', snOperator.FullyConnected(256), 'F2') \
   .addNode('F2', snOperator.FullyConnected(10), 'LS') \
   .addNode('LS', snOperator.LossFunction(snType.lossType.softMaxToCrossEntropy), 'Output')
   
   .............

# cycle lern
for n in range(1000):
   acc = [0]  
   net.training(lr, inLayer, outLayer, targLayer, acc)

   # calc accurate
   acc[0] = 0
   for i in range(bsz):
       if (np.argmax(outLayer[i][0][0]) == np.argmax(targLayer[i][0][0])):
           acc[0] += 1

   accuratSumm += acc[0]/bsz

   print(datetime.datetime.now().strftime('%H:%M:%S'), n, "accurate", accuratSumm / (n + 1))