MCPcopy
hub / github.com/shenweichen/DeepCTR-Torch / DCNMix

Class DCNMix

deepctr_torch/models/dcnmix.py:20–102  ·  view source on GitHub ↗

Instantiates the DCN-Mix model. :param linear_feature_columns: An iterable containing all the features used by linear part of the model. :param dnn_feature_columns: An iterable containing all the features used by deep part of the model. :param cross_num: positive integet,cross layer num

Source from the content-addressed store, hash-verified

18
19
20class DCNMix(BaseModel):
21 """Instantiates the DCN-Mix model.
22
23 :param linear_feature_columns: An iterable containing all the features used by linear part of the model.
24 :param dnn_feature_columns: An iterable containing all the features used by deep part of the model.
25 :param cross_num: positive integet,cross layer number
26 :param dnn_hidden_units: list,list of positive integer or empty list, the layer number and units in each layer of DNN
27 :param l2_reg_embedding: float. L2 regularizer strength applied to embedding vector
28 :param l2_reg_cross: float. L2 regularizer strength applied to cross net
29 :param l2_reg_dnn: float. L2 regularizer strength applied to DNN
30 :param init_std: float,to use as the initialize std of embedding vector
31 :param seed: integer ,to use as random seed.
32 :param dnn_dropout: float in [0,1), the probability we will drop out a given DNN coordinate.
33 :param dnn_use_bn: bool. Whether use BatchNormalization before activation or not DNN
34 :param dnn_activation: Activation function to use in DNN
35 :param low_rank: Positive integer, dimensionality of low-rank sapce.
36 :param num_experts: Positive integer, number of experts.
37 :param task: str, ``"binary"`` for binary logloss or ``"regression"`` for regression loss
38 :param device: str, ``"cpu"`` or ``"cuda:0"``
39 :param gpus: list of int or torch.device for multiple gpus. If None, run on `device`. `gpus[0]` should be the same gpu with `device`.
40 :return: A PyTorch model instance.
41
42 """
43
44 def __init__(self, linear_feature_columns,
45 dnn_feature_columns, cross_num=2,
46 dnn_hidden_units=(128, 128), l2_reg_linear=0.00001,
47 l2_reg_embedding=0.00001, l2_reg_cross=0.00001, l2_reg_dnn=0, init_std=0.0001, seed=1024,
48 dnn_dropout=0, low_rank=32, num_experts=4,
49 dnn_activation='relu', dnn_use_bn=False, task='binary', device='cpu', gpus=None):
50
51 super(DCNMix, self).__init__(linear_feature_columns=linear_feature_columns,
52 dnn_feature_columns=dnn_feature_columns, l2_reg_embedding=l2_reg_embedding,
53 init_std=init_std, seed=seed, task=task, device=device, gpus=gpus)
54 self.dnn_hidden_units = dnn_hidden_units
55 self.cross_num = cross_num
56 self.dnn = DNN(self.compute_input_dim(dnn_feature_columns), dnn_hidden_units,
57 activation=dnn_activation, use_bn=dnn_use_bn, l2_reg=l2_reg_dnn, dropout_rate=dnn_dropout,
58 init_std=init_std, device=device)
59 if len(self.dnn_hidden_units) > 0 and self.cross_num > 0:
60 dnn_linear_in_feature = self.compute_input_dim(dnn_feature_columns) + dnn_hidden_units[-1]
61 elif len(self.dnn_hidden_units) > 0:
62 dnn_linear_in_feature = dnn_hidden_units[-1]
63 elif self.cross_num > 0:
64 dnn_linear_in_feature = self.compute_input_dim(dnn_feature_columns)
65
66 self.dnn_linear = nn.Linear(dnn_linear_in_feature, 1, bias=False).to(
67 device)
68 self.crossnet = CrossNetMix(in_features=self.compute_input_dim(dnn_feature_columns),
69 low_rank=low_rank, num_experts=num_experts,
70 layer_num=cross_num, device=device)
71 self.add_regularization_weight(
72 filter(lambda x: 'weight' in x[0] and 'bn' not in x[0], self.dnn.named_parameters()), l2=l2_reg_dnn)
73 self.add_regularization_weight(self.dnn_linear.weight, l2=l2_reg_linear)
74 regularization_modules = [self.crossnet.U_list, self.crossnet.V_list, self.crossnet.C_list]
75 for module in regularization_modules:
76 self.add_regularization_weight(module, l2=l2_reg_cross)
77

Callers 1

test_DCNMixFunction · 0.90

Calls

no outgoing calls

Tested by 1

test_DCNMixFunction · 0.72