内容简介:MXNet 实现 Self-normalizing networks
MXNET-Scala Self-Norm Nets
MXNet-scala module implementation of Self-normalizing networks[1].
Based on: https://github.com/bioinf-jku/SNNs
Building
Tested on Ubuntu 14.04
Requirements
- sbt 0.13
- Mxnet
steps
1, compile Mxnet with CUDA, then compile the scala-pkg;
2,
cd Mxnet-Scala/SelfNormNets mkdir lib
3, copy your compiled mxnet-full_2.11-linux-x86_64-gpu-0.10.1-SNAPSHOT.jar into lib folder;
4, run sbt, compile the project
Running
1, use datas/get_mnist_data.sh script to download the mnist dataset
2, run trainSNN_CNN_MNIST.sh
or trainSNN_MLP_MNIST.sh
under scripts folder
Training logs
Train MLP
logs
bash trainSNN_MLP_MNIST.sh Epoch[0] Train-accuracy=0.86191666 Epoch[0] Time cost=1646 Epoch[0] Validation-accuracy=0.9341 Epoch[1] Train-accuracy=0.9213667 Epoch[1] Time cost=1478 Epoch[1] Validation-accuracy=0.9485 Epoch[2] Train-accuracy=0.9421667 Epoch[2] Time cost=1428 Epoch[2] Validation-accuracy=0.9402 Epoch[3] Train-accuracy=0.9501 Epoch[3] Time cost=1415 Epoch[3] Validation-accuracy=0.9669 Epoch[4] Train-accuracy=0.9571667 Epoch[4] Time cost=1604 Epoch[4] Validation-accuracy=0.9623 Epoch[5] Train-accuracy=0.96195 Epoch[5] Time cost=1457 Epoch[5] Validation-accuracy=0.9614 Epoch[6] Train-accuracy=0.9679667 Epoch[6] Time cost=1591 Epoch[6] Validation-accuracy=0.9673 Epoch[7] Train-accuracy=0.97048336 Epoch[7] Time cost=1629 Epoch[7] Validation-accuracy=0.9639 Epoch[8] Train-accuracy=0.9719333 Epoch[8] Time cost=1668 Epoch[8] Validation-accuracy=0.9703 Epoch[9] Train-accuracy=0.9753 Epoch[9] Time cost=1662 Epoch[9] Validation-accuracy=0.9728 Epoch[10] Train-accuracy=0.9769 Epoch[10] Time cost=1526 Epoch[10] Validation-accuracy=0.9752 Epoch[11] Train-accuracy=0.9784333 Epoch[11] Time cost=1487 Epoch[11] Validation-accuracy=0.9709 Epoch[12] Train-accuracy=0.98066664 Epoch[12] Time cost=1609 Epoch[12] Validation-accuracy=0.9753 Epoch[13] Train-accuracy=0.98113334 Epoch[13] Time cost=1475 Epoch[13] Validation-accuracy=0.9725 Epoch[14] Train-accuracy=0.98215 Epoch[14] Time cost=1477 Epoch[14] Validation-accuracy=0.9749
Compare selu with relu
logs
bash trainSNN_CNN_MNIST.sh Epoch[0] SNN Train-accuracy=0.88266224 Epoch[0] ReLU Train-accuracy=0.807926 Epoch[1] SNN Train-accuracy=0.9415899 Epoch[1] ReLU Train-accuracy=0.8241854 Epoch[2] SNN Train-accuracy=0.95097154 Epoch[2] ReLU Train-accuracy=0.8243189 Epoch[3] SNN Train-accuracy=0.95880073 Epoch[3] ReLU Train-accuracy=0.833734 Epoch[4] SNN Train-accuracy=0.9629741 Epoch[4] ReLU Train-accuracy=0.82568777 Epoch[5] SNN Train-accuracy=0.96793205 Epoch[5] ReLU Train-accuracy=0.8318643 Epoch[6] SNN Train-accuracy=0.9703693 Epoch[6] ReLU Train-accuracy=0.8342181 Epoch[7] SNN Train-accuracy=0.97163796 Epoch[7] ReLU Train-accuracy=0.83628803 Epoch[8] SNN Train-accuracy=0.9741086 Epoch[8] ReLU Train-accuracy=0.8316807 Epoch[9] SNN Train-accuracy=0.9753105 Epoch[9] ReLU Train-accuracy=0.8397269 SNN Validation-accuracy=0.96334136 ReLU Validation-accuracy=0.9423077
Referneces
[1] Klambauer, Günter, et al. "Self-Normalizing Neural Networks." arXiv preprint arXiv:1706.02515 (2017).
以上就是本文的全部内容,希望本文的内容对大家的学习或者工作能带来一定的帮助,也希望大家多多支持 码农网
猜你喜欢:- php如何实现session,自己实现session,laravel如何实现session
- AOP如何实现及实现原理
- webpack 实现 HMR 及其实现原理
- Docker实现原理之 - OverlayFS实现原理
- 为什么实现 .NET 的 ICollection 集合时需要实现 SyncRoot 属性?如何正确实现这个属性?
- 自己实现集合框架(十):顺序栈的实现
本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们。
信息检索导论
Christopher D.Manning、Hinrich Schütze、Prabhakar Raghavan / 王斌 / 人民邮电出版社 / 201008 / 69.00元
封面图片为英国伯明翰塞尔福瑞吉百货大楼,其极具线条感的轮廓外型优美,犹如水波的流动。其外表悬挂了1.5万个铝碟,创造出一种极具现代气息的纹理装饰效果,有如夜空下水流的波光粼粼,闪烁于月光之下,使建筑的商业氛围表现到极致。设计该建筑的英国“未来系统建筑事物所”,将商场内部围合成一个顶部采光的中庭,配以交叉的自动扶梯,使购物环境呈现出一种凝聚的向心力和商业广告的展示效应。作为英国第二商业城市伯明翰的建......一起来看看 《信息检索导论》 这本书的介绍吧!