Instantiates the DeepFM Network architecture. :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 use_fm: bool,use FM part or n
| 14 | |
| 15 | |
| 16 | class DeepFM(BaseModel): |
| 17 | """Instantiates the DeepFM Network architecture. |
| 18 | |
| 19 | :param linear_feature_columns: An iterable containing all the features used by linear part of the model. |
| 20 | :param dnn_feature_columns: An iterable containing all the features used by deep part of the model. |
| 21 | :param use_fm: bool,use FM part or not |
| 22 | :param dnn_hidden_units: list,list of positive integer or empty list, the layer number and units in each layer of DNN |
| 23 | :param l2_reg_linear: float. L2 regularizer strength applied to linear part |
| 24 | :param l2_reg_embedding: float. L2 regularizer strength applied to embedding vector |
| 25 | :param l2_reg_dnn: float. L2 regularizer strength applied to DNN |
| 26 | :param init_std: float,to use as the initialize std of embedding vector |
| 27 | :param seed: integer ,to use as random seed. |
| 28 | :param dnn_dropout: float in [0,1), the probability we will drop out a given DNN coordinate. |
| 29 | :param dnn_activation: Activation function to use in DNN |
| 30 | :param dnn_use_bn: bool. Whether use BatchNormalization before activation or not in DNN |
| 31 | :param task: str, ``"binary"`` for binary logloss or ``"regression"`` for regression loss |
| 32 | :param device: str, ``"cpu"`` or ``"cuda:0"`` |
| 33 | :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`. |
| 34 | :return: A PyTorch model instance. |
| 35 | |
| 36 | """ |
| 37 | |
| 38 | def __init__(self, |
| 39 | linear_feature_columns, dnn_feature_columns, use_fm=True, |
| 40 | dnn_hidden_units=(256, 128), |
| 41 | l2_reg_linear=0.00001, l2_reg_embedding=0.00001, l2_reg_dnn=0, init_std=0.0001, seed=1024, |
| 42 | dnn_dropout=0, |
| 43 | dnn_activation='relu', dnn_use_bn=False, task='binary', device='cpu', gpus=None): |
| 44 | |
| 45 | super(DeepFM, self).__init__(linear_feature_columns, dnn_feature_columns, l2_reg_linear=l2_reg_linear, |
| 46 | l2_reg_embedding=l2_reg_embedding, init_std=init_std, seed=seed, task=task, |
| 47 | device=device, gpus=gpus) |
| 48 | |
| 49 | self.use_fm = use_fm |
| 50 | self.use_dnn = len(dnn_feature_columns) > 0 and len( |
| 51 | dnn_hidden_units) > 0 |
| 52 | if use_fm: |
| 53 | self.fm = FM() |
| 54 | |
| 55 | if self.use_dnn: |
| 56 | self.dnn = DNN(self.compute_input_dim(dnn_feature_columns), dnn_hidden_units, |
| 57 | activation=dnn_activation, l2_reg=l2_reg_dnn, dropout_rate=dnn_dropout, use_bn=dnn_use_bn, |
| 58 | init_std=init_std, device=device) |
| 59 | self.dnn_linear = nn.Linear( |
| 60 | dnn_hidden_units[-1], 1, bias=False).to(device) |
| 61 | |
| 62 | self.add_regularization_weight( |
| 63 | filter(lambda x: 'weight' in x[0] and 'bn' not in x[0], self.dnn.named_parameters()), l2=l2_reg_dnn) |
| 64 | self.add_regularization_weight(self.dnn_linear.weight, l2=l2_reg_dnn) |
| 65 | self.to(device) |
| 66 | |
| 67 | def forward(self, X): |
| 68 | |
| 69 | sparse_embedding_list, dense_value_list = self.input_from_feature_columns(X, self.dnn_feature_columns, |
| 70 | self.embedding_dict) |
| 71 | logit = self.linear_model(X) |
| 72 | |
| 73 | if self.use_fm and len(sparse_embedding_list) > 0: |
no outgoing calls