Ejemplo n.º 1
0
def main(_):
    model_dir = os.path.join(os.path.dirname(__file__), 'logdata')
    print(model_dir)
    # Build 2 hidden layer DNN with 10, 10 units respectively.
    if True:
        ws = tf.estimator.WarmStartSettings(
            ckpt_to_initialize_from="/home/lucius/Projects/notebook/homework/ctr_predict/models/v3/logdata_95",
            vars_to_warm_start=".*input_layer.*")
    else:
        ws = None

    for i in range(15):
        classifier = tf.estimator.Estimator(
            model_fn=my_model,
            params={
                'feature_columns': build_model_columns_fm(),
                'optimizer': 'sgd',
                'learning_rate': 0.001/(i + 1),
            },
            model_dir=model_dir,
            warm_start_from=ws)

        print('perform {}s epoch...'.format(i))
        # Train the Model.
        classifier.train(
           input_fn=get_input_fn(os.path.join(os.path.dirname(__file__), '../../FE/FE1/train_split.tfrecord'), 256, 1, 5000, use_tfrecord=True))

        # Evaluate the model.
        eval_result = classifier.evaluate(
            input_fn=get_input_fn(os.path.join(os.path.dirname(__file__), '../../FE/FE1/valid_split.tfrecord'), 256, 1, 5000, use_tfrecord=True))

        print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))
Ejemplo n.º 2
0
def main(_):
    model_dir = os.path.join(os.path.dirname(__file__), 'logdata_fm_noreg')
    restore_map = {
        'input_layer/' + feature_name + '_embed_lr': feature_name + '_embed_lr'
        for feature_name in COLUMN_NAMES[2:]
    }
    restore_map.update({
        'input_layer/' + feature_name + '_embed_fm': feature_name + '_embed_lr'
        for feature_name in COLUMN_NAMES[2:]
    })
    # Build 2 hidden layer DNN with 10, 10 units respectively.
    if True:
        ws = tf.estimator.WarmStartSettings(
            ckpt_to_initialize_from=
            "/home/lucius/Projects/notebook/homework/ctr_predict/models/v2/logdata_3_4",
            vars_to_warm_start=".*input_layer.*")
    else:
        ws = None

    classifier = tf.estimator.Estimator(model_fn=my_model,
                                        params={
                                            'feature_columns':
                                            build_model_columns_fm(),
                                            'optimizer':
                                            'sgd'
                                        },
                                        model_dir=model_dir,
                                        warm_start_from=ws)

    for i in range(15):
        print('perform {}s epoch...'.format(i))
        # Train the Model.
        classifier.train(input_fn=get_input_fn(os.path.join(
            os.path.dirname(__file__), '../../FE/FE1/train_split.tfrecord'),
                                               256,
                                               1,
                                               5000,
                                               use_tfrecord=True))

        # Evaluate the model.
        eval_result = classifier.evaluate(input_fn=get_input_fn(
            os.path.join(os.path.dirname(__file__),
                         '../../FE/FE1/valid_split.tfrecord'),
            256,
            1,
            5000,
            use_tfrecord=True))

        print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))
Ejemplo n.º 3
0
def main(_):
    # Build 2 hidden layer DNN with 10, 10 units respectively.
    if False:
        ws = tf.estimator.WarmStartSettings(
            ckpt_to_initialize_from="/home/lucius/Projects/notebook/homework/ctr_predict/models/v1/logdata_fm_10",
            vars_to_warm_start=".*input_layer.*")
    else:
        ws = None

    for i in range(5):
        fm_dim = i*5 + 3
        for j in range(5):
            model_reg = (10**(-j)) * 1.0
            model_dir = os.path.join(os.path.dirname(__file__), 'logdata_' + str(i) + '_' + str(j))
            for k in range(1):
                classifier = tf.estimator.Estimator(
                    model_fn=my_model,
                    params={
                        'feature_columns': build_model_columns_fm(),
                        'optimizer': 'sgd' if k != 0 else None,
                        'learning_rate': 0.0001,
                        'dim': fm_dim,
                        'regularizer': model_reg
                    },
                    model_dir=model_dir,
                    warm_start_from=ws)

                print('perform {}s epoch...'.format(i))
                # Train the Model.
                # classifier.train(
                #    input_fn=get_input_fn(os.path.join(os.path.dirname(__file__), '../../FE/FE1/train_split.tfrecord'), 256, 1, 5000, use_tfrecord=True))

                # Evaluate the model.
                eval_result = classifier.evaluate(
                    input_fn=get_input_fn(os.path.join(os.path.dirname(__file__), '../../FE/FE1/valid_split.tfrecord'), 256, 1, 5000, use_tfrecord=True))

                print('\nTest set accuracy: {accuracy:0.3f}\n'.format(**eval_result))