- Autograd API: w.attach_grad()
- Callback API
- Optimization API 优化算法:待优化参数、优化算法、学习率等
- NDArray API: nd.random.normal
- Module API
- Image API
- IO API: Data Loading API
- KV Store API
- Metric API
- Profiler API
- Contrib Package
- Run-Time Compilation API
- Run-Time Feature detection / Library Info
- Symbol API
- Symbol in Pictures API
- Tools
- Gluon API
- Gluon Neural Network Layers: mxnet.gluon.nn.
- Basic、Convolutional、Pooling、Activation Layers 激活函数
- Gluon Recurrent Neural Network API: RNN mxnet.gluon.rnn.
- Gluon Loss API: 损失函数 mxnet.gluon.loss.
- Gluon Data API: mxnet.gluon.data.;gdata.DataLoader
- Gluon Model Zoo
- Gluon Contrib API
- Parameter: Parameter\Constant\ParameterDict
- Containers: Block\HybridBlock\SymbolBlock\ nn.Sequential \nn.HybridSequential
- Trainer: Trainer
- Utilities: split_data\split_and_load\clip_global_norm
训练模型自己写:epoch次数,data_iter迭代,如何record求导并backward,然后用Trainer实例按批次根据loss和学习率迭代更新参数,然后显式计算loss打印给人看
处理数据
建造模型
初始化参数
损失函数
优化算法
训练模型:Trainer(net、训练/训练数据集iter、迭代批次、优化函数、训练总周期、ctx)