NumpyFlow

详细文档请参照：https://blog.csdn.net/kid_14_12/article/details/105852626

待添加.....

Tensor

封装numpy数组，是NumpyFlow的数据载体，相当于torch中的Tensor。

已完成

基本类

• [x] Operation：用于支持基本的运算及对应的梯度计算，是支持自动微分的基本算子
• [x] Tensor：
• [x] Optimizer
• [x] Module

Operation

• [x] Assign、Add、Multiply、Subtract
• [x] Divide、Negative、Positive、Power
• [x] Exp、Log、Log2、Log10
• [x] MatMul、EinSum

Optimizer

• [x] sgd
• [ ] adam

初始化方法

• [x] Kaiming初始化
• [ ] Xavier初始化
• [ ] 随机初始化
• [x] 填充0初始化
• [x] 填充1初始化

Module

核心网络层

• [x] Linear
• [x] Relu、Sigmid、Softmax
• [ ] Conv1D
• [x] Conv2D
• [ ] Conv3D
• [ ] MaxPool1D
• [x] MaxPool2D
• [ ] MaxPool3D
• [ ] BatchNorm1D
• [x] BatchNorm2D
• [ ] BatchNorm3D
• [ ] Reshape
• [ ] Permute
• [ ] Flatten
• [ ] RepeatVector
• [ ] Lambda
• [ ] ActivityRegularization
• [ ] Masking
• [ ] SpatialDropout1D
• [ ] SpatialDropout2D
• [ ] SpatialDropout3D

局部连接层和循环层

• [ ] LocallyConnected1D
• [ ] LocallyConnected2D
• [ ] RNN
• [ ] GRU
• [ ] LSTM
• [ ] ConvLSTM2D
• [ ] SimpleRNNCell
• [ ] GRUCell
• [ ] LSTMCell
• [ ] CuDNNGRU
• [ ] CuDNNLSTM
• [ ] Embedding