deepctr_torch.models.dien module

Author:

Ze Wang, wangze0801@126.com

Reference:

[1] Zhou G, Mou N, Fan Y, et al. Deep Interest Evolution Network for Click-Through Rate Prediction[J]. arXiv preprint arXiv:1809.03672, 2018. (https://arxiv.org/pdf/1809.03672.pdf)

class deepctr_torch.models.dien.DIEN(dnn_feature_columns, history_feature_list, gru_type='GRU', use_negsampling=False, alpha=1.0, use_bn=False, dnn_hidden_units=(256, 128), dnn_activation='relu', att_hidden_units=(64, 16), att_activation='relu', att_weight_normalization=True, l2_reg_dnn=0, l2_reg_embedding=1e-06, dnn_dropout=0, init_std=0.0001, seed=1024, task='binary', device='cpu', gpus=None)

Instantiates the Deep Interest Evolution Network architecture.

Parameters:

• dnn_feature_columns – An iterable containing all the features used by deep part of the model.
• history_feature_list – list,to indicate sequence sparse field
• gru_type – str,can be GRU AIGRU AUGRU AGRU
• use_negsampling – bool, whether or not use negtive sampling
• alpha – float ,weight of auxiliary_loss
• use_bn – bool. Whether use BatchNormalization before activation or not in deep net
• dnn_hidden_units – list,list of positive integer or empty list, the layer number and units in each layer of DNN
• dnn_activation – Activation function to use in DNN
• att_hidden_units – list,list of positive integer , the layer number and units in each layer of attention net
• att_activation – Activation function to use in attention net
• att_weight_normalization – bool.Whether normalize the attention score of local activation unit.
• l2_reg_dnn – float. L2 regularizer strength applied to DNN
• l2_reg_embedding – float. L2 regularizer strength applied to embedding vector
• dnn_dropout – float in [0,1), the probability we will drop out a given DNN coordinate.
• init_std – float,to use as the initialize std of embedding vector
• seed – integer ,to use as random seed.
• task – str, "binary" for binary logloss or "regression" for regression loss
• 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)

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.

class deepctr_torch.models.dien.InterestEvolving(input_size, gru_type='GRU', use_neg=False, init_std=0.001, att_hidden_size=(64, 16), att_activation='sigmoid', att_weight_normalization=False)

forward(query, keys, keys_length, mask=None)

query: 2D tensor, [B, H] keys: (masked_interests), 3D tensor, [b, T, H] keys_length: 1D tensor, [B]

outputs: 2D tensor, [B, H]

class deepctr_torch.models.dien.InterestExtractor(input_size, use_neg=False, init_std=0.001, device='cpu')

forward(keys, keys_length, neg_keys=None)

keys: 3D tensor, [B, T, H] keys_length: 1D tensor, [B] neg_keys: 3D tensor, [B, T, H]

masked_interests: 2D tensor, [b, H] aux_loss: [1]