|
| 1 | +# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved. |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | + |
| 15 | +import math |
| 16 | +import paddle |
| 17 | + |
| 18 | +from net import NAMLLayer |
| 19 | + |
| 20 | + |
| 21 | +class StaticModel(): |
| 22 | + def __init__(self, config): |
| 23 | + self.cost = None |
| 24 | + self.infer_target_var = None |
| 25 | + self.config = config |
| 26 | + self._init_hyper_parameters() |
| 27 | + |
| 28 | + def _init_hyper_parameters(self): |
| 29 | + self.article_content_size = self.config.get( |
| 30 | + "hyper_parameters.article_content_size") |
| 31 | + self.article_title_size = self.config.get( |
| 32 | + "hyper_parameters.article_title_size") |
| 33 | + self.browse_size = self.config.get("hyper_parameters.browse_size") |
| 34 | + self.neg_condidate_sample_size = self.config.get( |
| 35 | + "hyper_parameters.neg_condidate_sample_size") |
| 36 | + self.word_dimension = self.config.get( |
| 37 | + "hyper_parameters.word_dimension") |
| 38 | + self.category_size = self.config.get("hyper_parameters.category_size") |
| 39 | + self.sub_category_size = self.config.get( |
| 40 | + "hyper_parameters.sub_category_size") |
| 41 | + self.cate_dimension = self.config.get( |
| 42 | + "hyper_parameters.category_dimension") |
| 43 | + self.word_dict_size = self.config.get( |
| 44 | + "hyper_parameters.word_dict_size") |
| 45 | + self.learning_rate = self.config.get( |
| 46 | + "hyper_parameters.optimizer.learning_rate") |
| 47 | + self.sample_size = self.neg_condidate_sample_size + 1 |
| 48 | + |
| 49 | + def create_feeds(self, is_infer=False): |
| 50 | + inputs = [ |
| 51 | + paddle.static.data( |
| 52 | + name="sampe_cate", |
| 53 | + shape=[None, self.sample_size], |
| 54 | + dtype='int64'), paddle.static.data( |
| 55 | + name="browse_cate", |
| 56 | + shape=[None, self.browse_size], |
| 57 | + dtype='int64'), paddle.static.data( |
| 58 | + name="sampe_sub_cate", |
| 59 | + shape=[None, self.sample_size], |
| 60 | + dtype='int64'), paddle.static.data( |
| 61 | + name="browse_sub_cate", |
| 62 | + shape=[None, self.browse_size], |
| 63 | + dtype='int64'), |
| 64 | + paddle.static.data( |
| 65 | + name="sampe_title", |
| 66 | + shape=[None, self.sample_size, self.article_title_size], |
| 67 | + dtype='int64'), paddle.static.data( |
| 68 | + name="browse_title", |
| 69 | + shape=[None, self.browse_size, self.article_title_size], |
| 70 | + dtype='int64'), |
| 71 | + paddle.static.data( |
| 72 | + name="sample_article", |
| 73 | + shape=[None, self.sample_size, self.article_content_size], |
| 74 | + dtype='int64'), paddle.static.data( |
| 75 | + name="browse_article", |
| 76 | + shape=[None, self.browse_size, self.article_content_size], |
| 77 | + dtype='int64') |
| 78 | + ] |
| 79 | + label = paddle.static.data( |
| 80 | + name="label", shape=[None, self.sample_size], dtype="int64") |
| 81 | + return [label] + inputs |
| 82 | + |
| 83 | + def net(self, input, is_infer=False): |
| 84 | + self.labels = input[0] |
| 85 | + self.sparse_inputs = input[1:] |
| 86 | + #self.dense_input = input[-1] |
| 87 | + #sparse_number = self.sparse_inputs_slots - 1 |
| 88 | + model = NAMLLayer(self.article_content_size, self.article_title_size, |
| 89 | + self.browse_size, self.neg_condidate_sample_size, |
| 90 | + self.word_dimension, self.category_size, |
| 91 | + self.sub_category_size, self.cate_dimension, |
| 92 | + self.word_dict_size) |
| 93 | + |
| 94 | + raw = model(self.sparse_inputs) |
| 95 | + |
| 96 | + soft_predict = paddle.nn.functional.sigmoid( |
| 97 | + paddle.reshape(raw, [-1, 1])) |
| 98 | + predict_2d = paddle.concat(x=[1 - soft_predict, soft_predict], axis=-1) |
| 99 | + labels = paddle.reshape(self.labels, [-1, 1]) |
| 100 | + #metrics_list[0].update(preds=predict_2d.numpy(), labels=labels.numpy()) |
| 101 | + #self.predict = predict_2d |
| 102 | + |
| 103 | + auc, batch_auc, _ = paddle.static.auc(input=predict_2d, |
| 104 | + label=labels, |
| 105 | + num_thresholds=2**12, |
| 106 | + slide_steps=20) |
| 107 | + self.inference_target_var = auc |
| 108 | + if is_infer: |
| 109 | + fetch_dict = {'auc': auc} |
| 110 | + return fetch_dict |
| 111 | + |
| 112 | + cost = paddle.nn.functional.cross_entropy( |
| 113 | + input=raw, |
| 114 | + label=paddle.cast(self.labels, "float32"), |
| 115 | + soft_label=True) |
| 116 | + avg_cost = paddle.mean(x=cost) |
| 117 | + self._cost = avg_cost |
| 118 | + |
| 119 | + fetch_dict = {'cost': avg_cost, 'auc': auc} |
| 120 | + return fetch_dict |
| 121 | + |
| 122 | + def create_optimizer(self, strategy=None): |
| 123 | + optimizer = paddle.optimizer.Adam( |
| 124 | + learning_rate=self.learning_rate, lazy_mode=True) |
| 125 | + if strategy != None: |
| 126 | + import paddle.distributed.fleet as fleet |
| 127 | + optimizer = fleet.distributed_optimizer(optimizer, strategy) |
| 128 | + optimizer.minimize(self._cost) |
| 129 | + |
| 130 | + def infer_net(self, input): |
| 131 | + return self.net(input, is_infer=True) |
0 commit comments