deepctr_torch.models.multitask.mmoe module

Author:
zanshuxun, zanshuxun@aliyun.com
Reference:
[1] Jiaqi Ma, Zhe Zhao, Xinyang Yi, et al. Modeling Task Relationships in Multi-task Learning with Multi-gate Mixture-of-Experts[C] (https://dl.acm.org/doi/10.1145/3219819.3220007)
class deepctr_torch.models.multitask.mmoe.MMOE(dnn_feature_columns, num_experts=3, expert_dnn_hidden_units=(256, 128), gate_dnn_hidden_units=(64, ), tower_dnn_hidden_units=(64, ), l2_reg_linear=1e-05, l2_reg_embedding=1e-05, l2_reg_dnn=0, init_std=0.0001, seed=1024, dnn_dropout=0, dnn_activation='relu', dnn_use_bn=False, task_types=('binary', 'binary'), task_names=('ctr', 'ctcvr'), device='cpu', gpus=None)[source]

Instantiates the Multi-gate Mixture-of-Experts architecture.

Parameters:
  • dnn_feature_columns – An iterable containing all the features used by deep part of the model.
  • num_experts – integer, number of experts.
  • expert_dnn_hidden_units – list, list of positive integer or empty list, the layer number and units in each layer of expert DNN.
  • gate_dnn_hidden_units – list, list of positive integer or empty list, the layer number and units in each layer of gate DNN.
  • tower_dnn_hidden_units – list, list of positive integer or empty list, the layer number and units in each layer of task-specific DNN.
  • l2_reg_linear – float, L2 regularizer strength applied to linear part.
  • l2_reg_embedding – float, L2 regularizer strength applied to embedding vector.
  • l2_reg_dnn – float, L2 regularizer strength applied to DNN.
  • init_std – float, to use as the initialize std of embedding vector.
  • seed – integer, to use as random seed.
  • dnn_dropout – float in [0,1), the probability we will drop out a given DNN coordinate.
  • dnn_activation – Activation function to use in DNN.
  • dnn_use_bn – bool, Whether use BatchNormalization before activation or not in DNN.
  • task_types – list of str, indicating the loss of each tasks, "binary" for binary logloss, "regression" for regression loss. e.g. [‘binary’, ‘regression’].
  • task_names – list of str, indicating the predict target of each tasks.
  • device – str, "cpu" or "cuda:0".
  • gpus – list of int or torch.device for multiple gpus. If None, run on device. gpus[0] should be the same gpu with device.
Returns:

A PyTorch model instance.
forward(X)[source]

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.