Shared_Bottom.py

"""
Reference:
    [1] Caruana R. Multitask learning[J]. Machine learning, 1997.(http://reports-archive.adm.cs.cmu.edu/anon/1997/CMU-CS-97-203.pdf)
"""

import tensorflow as tf

from deepctr.feature_column import build_input_features, input_from_feature_columns
from deepctr.layers.core import PredictionLayer, DNN
from deepctr.layers.utils import combined_dnn_input

def Shared_Bottom(dnn_feature_columns, num_tasks=None, task_types=None, task_names=None,
                  bottom_dnn_units=[128, 128], tower_dnn_units_lists=[[64,32], [64,32]],
                  l2_reg_embedding=0.00001, l2_reg_dnn=0, seed=1024, dnn_dropout=0,dnn_activation='relu', dnn_use_bn=False):
    """Instantiates the Shared_Bottom multi-task learning Network architecture.

    :param dnn_feature_columns: An iterable containing all the features used by deep part of the model.
    :param num_tasks:  integer, number of tasks, equal to number of outputs, must be greater than 1.
    :param task_types: list of str, indicating the loss of each tasks, ``"binary"`` for  binary logloss or  ``"regression"`` for regression loss. e.g. ['binary', 'regression']
    :param task_names: list of str, indicating the predict target of each tasks

    :param bottom_dnn_units: list,list of positive integer or empty list, the layer number and units in each layer of shared-bottom DNN
    :param tower_dnn_units_lists: list, list of positive integer list, its length must be euqal to num_tasks, the layer number and units in each layer of task-specific DNN

    :param l2_reg_embedding: float. L2 regularizer strength applied to embedding vector
    :param l2_reg_dnn: float. L2 regularizer strength applied to DNN
    :param seed: integer ,to use as random seed.
    :param dnn_dropout: float in [0,1), the probability we will drop out a given DNN coordinate.
    :param dnn_activation: Activation function to use in DNN
    :param dnn_use_bn: bool. Whether use BatchNormalization before activation or not in DNN
    :return: A Keras model instance.
    """
    if num_tasks <= 1:
        raise ValueError("num_tasks must be greater than 1")
    if len(task_types) != num_tasks:
        raise ValueError("num_tasks must be equal to the length of task_types")

    for task_type in task_types:
        if task_type not in ['binary', 'regression']:
            raise ValueError("task must be binary or regression, {} is illegal".format(task_type))

    if num_tasks != len(tower_dnn_units_lists):
        raise ValueError("the length of tower_dnn_units_lists must be euqal to num_tasks")

    features = build_input_features(dnn_feature_columns)
    inputs_list = list(features.values())

    sparse_embedding_list, dense_value_list = input_from_feature_columns(features, dnn_feature_columns, l2_reg_embedding,seed)

    dnn_input = combined_dnn_input(sparse_embedding_list, dense_value_list)
    shared_bottom_output = DNN(bottom_dnn_units, dnn_activation, l2_reg_dnn, dnn_dropout, dnn_use_bn, seed=seed)(dnn_input)

    tasks_output = []
    for task_type, task_name, tower_dnn in zip(task_types, task_names, tower_dnn_units_lists):
        tower_output = DNN(tower_dnn, dnn_activation, l2_reg_dnn, dnn_dropout, dnn_use_bn, seed=seed, name='tower_'+task_name)(shared_bottom_output)

        logit = tf.keras.layers.Dense(1, use_bias=False, activation=None)(tower_output)
        output = PredictionLayer(task_type, name=task_name)(logit) #regression->keep, binary classification->sigmoid
        tasks_output.append(output)

    model = tf.keras.models.Model(inputs=inputs_list, outputs=tasks_output)
    return model


if __name__ == "__main__":
    from utils import get_mtl_data
    dnn_feature_columns, train_model_input, test_model_input, y_list = get_mtl_data()
    model = Shared_Bottom(dnn_feature_columns, num_tasks=2, task_types= ['binary', 'binary'],
                          task_names=['label_income','label_marital'], bottom_dnn_units=[16],
                          tower_dnn_units_lists=[[8],[8]])
    model.compile("adam", loss=["binary_crossentropy", "binary_crossentropy"], metrics=['AUC'])
    history = model.fit(train_model_input, y_list, batch_size=256, epochs=5, verbose=2, validation_split=0.0 )