##ABOUT This is a mini deep learning framework created by xiaohaipeng @tianjin and changsha. The framework is built by numpy library,and initially realized the construction of MLP and LSTM networks,which can be used to solve regression and classification problems
##Features
- Define Dynamic Computing Graph
- Define Neural Network Operator
- Define Placeholder,Linear,Sigmoid,Relu,Leakrelu,Elu,Tanh,LSTM,MSE,EntropyCrossLossWithSoftmax.
- Auto-Diff Computing
- Auto-Feedforward and Backward
- Save and Load model
- Examples for using MLP,LSTM in Regression and classification problem
- A algorithm is designed to prevent gradient explosion and gradient disappearance
#How to USE 1.You can install xhp_flow by pip3 install xhp_flow and run the python file in examples folder. 2.Also you can run python flie directly.
#Remain
- LSTM Classification effect is not good enough,I find the the cause is gradient disappearance(the gradient of LSTM is close to zero),so I write a function Visual_gradient() to see the gradient.So far,I have not found a good solution,I just relieved it a litle bit,but the classification effect is still not good.
- CNN
- SGD,Adam....
Connection
wechat:18651271973 Mail:307153814@qq.com