本文件根据../textclf/config/classifier.py自动生成
ClassifierConfig有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| input_size | Optional[int] | None | input_size是embedding layer层的维度 |
| output_size | Optional[int] | None | output_size是输出空间的大小,应该与标签数量相对应 |
CNNClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| kernel_sizes | List[int] | [2, 3, 4] | |
| num_kernels | int | 256 | |
| top_k_max_pooling | int | 1 # max top-k pooling. | |
| hidden_layer_dropout | float | 0.5 |
LinearClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| pool_method | str | "first" | first or meanif first, use first token's embedidng as Linear layer inputif mean, average sequences embedding as linear layer input |
| embedding_dropout | float | 0.1 | dropout probability of the embedding layer |
| hidden_units | Optional[List[int]] | None | 隐藏层设置 |
| activations | Optional[List[str]] | None | |
| hidden_dropouts | Optional[List[float]] | None |
RNNClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| rnn_config | RNNConfig | RNNConfig() | config for RNN layer |
| use_attention | bool | True | If True, use attention mechanism to caluate output state |
| dropout | float | 0.2 | dropout probability on context |
RCNNClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| rnn_config | RNNConfig | RNNConfig() | config for RNN layer |
| semantic_units | int | 512 | size of latent semantic vectorRefer to Equation 4 of the original paper"Recurrent Convolutional Neural Networks for Text Classification" |
DRNNClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| rnn_config | RNNConfig | RNNConfig() | config for RNN layer |
| dropout | float | 0.2 | The dropout probability applied in drnninput and output layers, |
| window_size | int | 10 | The window size for rnn |
DPCNNClassifierConfig继承ClassifierConfig的所有属性，同时它还有以下属性：
| Attribute name | Type | Default | Description |
|---|---|---|---|
| kernel_size | int | 3 | kernel size. |
| pooling_stride | int | 2 | stride of pooling. |
| num_kernels | int | 16 | number of kernels. |
| blocks | int | 2 | number of blocks for DPCNN. |
| dropout | float | 0.2 | dropout probability on convolution features |