model.py 3.7 KB
Newer Older
T
tangwei 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
# Copyright (c) 2020 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

T
tangwei 已提交
15
import math
T
tangwei 已提交
16

T
tangwei 已提交
17
import paddle.fluid as fluid
T
tangwei 已提交
18

19 20
from paddlerec.core.utils import envs
from paddlerec.core.model import Model as ModelBase
T
tangwei 已提交
21 22


T
tangwei 已提交
23
class Model(ModelBase):
T
tangwei 已提交
24
    def __init__(self, config):
T
tangwei 已提交
25
        ModelBase.__init__(self, config)
T
tangwei 已提交
26

27 28 29
    def _init_hyper_parameters(self):
        self.is_distributed = True if envs.get_trainer(
        ) == "CtrTrainer" else False
X
fix  
xjqbest 已提交
30 31 32 33 34
        self.sparse_feature_number = envs.get_global_env(
            "hyper_parameters.sparse_feature_number")
        self.sparse_feature_dim = envs.get_global_env(
            "hyper_parameters.sparse_feature_dim")
        self.learning_rate = envs.get_global_env(
X
test  
xjqbest 已提交
35
            "hyper_parameters.optimizer.learning_rate")
36 37

    def net(self, input, is_infer=False):
X
xujiaqi01 已提交
38
        self.sparse_inputs = self._sparse_data_var[1:]
X
xjqbest 已提交
39
        self.dense_input = self._dense_data_var[0]
X
xujiaqi01 已提交
40
        self.label_input = self._sparse_data_var[0]
T
tangwei 已提交
41

T
tangwei 已提交
42
        def embedding_layer(input):
T
tangwei 已提交
43 44 45
            emb = fluid.layers.embedding(
                input=input,
                is_sparse=True,
46 47
                is_distributed=self.is_distributed,
                size=[self.sparse_feature_number, self.sparse_feature_dim],
T
tangwei 已提交
48 49
                param_attr=fluid.ParamAttr(
                    name="SparseFeatFactors",
T
tangwei 已提交
50 51
                    initializer=fluid.initializer.Uniform()), )
            emb_sum = fluid.layers.sequence_pool(input=emb, pool_type='sum')
T
tangwei 已提交
52 53 54
            return emb_sum

        sparse_embed_seq = list(map(embedding_layer, self.sparse_inputs))
X
xjqbest 已提交
55 56
        concated = fluid.layers.concat(
            sparse_embed_seq + [self.dense_input], axis=1)
T
tangwei 已提交
57 58

        fcs = [concated]
X
fix  
xjqbest 已提交
59
        hidden_layers = envs.get_global_env("hyper_parameters.fc_sizes")
T
tangwei 已提交
60 61

        for size in hidden_layers:
62 63 64 65 66 67 68 69
            output = fluid.layers.fc(
                input=fcs[-1],
                size=size,
                act='relu',
                param_attr=fluid.ParamAttr(
                    initializer=fluid.initializer.Normal(
                        scale=1.0 / math.sqrt(fcs[-1].shape[1]))))
            fcs.append(output)
T
tangwei 已提交
70 71 72 73 74 75

        predict = fluid.layers.fc(
            input=fcs[-1],
            size=2,
            act="softmax",
            param_attr=fluid.ParamAttr(initializer=fluid.initializer.Normal(
T
tangwei 已提交
76
                scale=1 / math.sqrt(fcs[-1].shape[1]))))
T
tangwei 已提交
77 78 79 80 81

        self.predict = predict

        auc, batch_auc, _ = fluid.layers.auc(input=self.predict,
                                             label=self.label_input,
T
tangwei 已提交
82
                                             num_thresholds=2**12,
T
tangwei 已提交
83
                                             slide_steps=20)
X
fix  
xjqbest 已提交
84 85 86 87 88
        if is_infer:
            self._infer_results["AUC"] = auc
            self._infer_results["BATCH_AUC"] = batch_auc
            return

T
tangwei 已提交
89 90
        self._metrics["AUC"] = auc
        self._metrics["BATCH_AUC"] = batch_auc
X
fix  
xjqbest 已提交
91 92 93 94
        cost = fluid.layers.cross_entropy(
            input=self.predict, label=self.label_input)
        avg_cost = fluid.layers.reduce_mean(cost)
        self._cost = avg_cost
T
tangwei12 已提交
95

T
tangwei 已提交
96
    def optimizer(self):
97
        optimizer = fluid.optimizer.Adam(self.learning_rate, lazy_mode=True)
T
tangwei 已提交
98 99
        return optimizer

T
tangwei 已提交
100
    def infer_net(self):
101
        pass