1 # Copyright 2017 The TensorFlow 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 """GTFlow Estimator definition.""" 16 17 from __future__ import absolute_import 18 from __future__ import division 19 from __future__ import print_function 20 21 from tensorflow.contrib.boosted_trees.estimator_batch import model 22 from tensorflow.contrib.boosted_trees.python.utils import losses 23 from tensorflow.contrib.learn.python.learn.estimators import estimator 24 from tensorflow.contrib.learn.python.learn.estimators import head as head_lib 25 from tensorflow.python.ops import math_ops 26 27 28 class GradientBoostedDecisionTreeClassifier(estimator.Estimator): 29 """An estimator using gradient boosted decision trees.""" 30 31 def __init__(self, 32 learner_config, 33 examples_per_layer, 34 n_classes=2, 35 num_trees=None, 36 feature_columns=None, 37 weight_column_name=None, 38 model_dir=None, 39 config=None, 40 label_keys=None, 41 feature_engineering_fn=None, 42 logits_modifier_function=None, 43 center_bias=True): 44 """Initializes a GradientBoostedDecisionTreeClassifier estimator instance. 45 46 Args: 47 learner_config: A config for the learner. 48 examples_per_layer: Number of examples to accumulate before growing a 49 layer. It can also be a function that computes the number of examples 50 based on the depth of the layer that's being built. 51 n_classes: Number of classes in the classification. 52 num_trees: An int, number of trees to build. 53 feature_columns: A list of feature columns. 54 weight_column_name: Name of the column for weights, or None if not 55 weighted. 56 model_dir: Directory for model exports, etc. 57 config: `RunConfig` object to configure the runtime settings. 58 label_keys: Optional list of strings with size `[n_classes]` defining the 59 label vocabulary. Only supported for `n_classes` > 2. 60 feature_engineering_fn: Feature engineering function. Takes features and 61 labels which are the output of `input_fn` and returns features and 62 labels which will be fed into the model. 63 logits_modifier_function: A modifier function for the logits. 64 center_bias: Whether a separate tree should be created for first fitting 65 the bias. 66 67 Raises: 68 ValueError: If learner_config is not valid. 69 """ 70 if n_classes > 2: 71 # For multi-class classification, use our loss implementation that 72 # supports second order derivative. 73 def loss_fn(labels, logits, weights=None): 74 result = losses.per_example_maxent_loss( 75 labels=labels, logits=logits, weights=weights, 76 num_classes=n_classes) 77 return math_ops.reduce_mean(result[0]) 78 else: 79 loss_fn = None 80 head = head_lib.multi_class_head( 81 n_classes=n_classes, 82 weight_column_name=weight_column_name, 83 enable_centered_bias=False, 84 loss_fn=loss_fn) 85 if learner_config.num_classes == 0: 86 learner_config.num_classes = n_classes 87 elif learner_config.num_classes != n_classes: 88 raise ValueError("n_classes (%d) doesn't match learner_config (%d)." % 89 (learner_config.num_classes, n_classes)) 90 super(GradientBoostedDecisionTreeClassifier, self).__init__( 91 model_fn=model.model_builder, 92 params={ 93 'head': head, 94 'feature_columns': feature_columns, 95 'learner_config': learner_config, 96 'num_trees': num_trees, 97 'weight_column_name': weight_column_name, 98 'examples_per_layer': examples_per_layer, 99 'center_bias': center_bias, 100 'logits_modifier_function': logits_modifier_function, 101 }, 102 model_dir=model_dir, 103 config=config, 104 feature_engineering_fn=feature_engineering_fn) 105 106 107 class GradientBoostedDecisionTreeRegressor(estimator.Estimator): 108 """An estimator using gradient boosted decision trees.""" 109 110 def __init__(self, 111 learner_config, 112 examples_per_layer, 113 label_dimension=1, 114 num_trees=None, 115 feature_columns=None, 116 label_name=None, 117 weight_column_name=None, 118 model_dir=None, 119 config=None, 120 feature_engineering_fn=None, 121 logits_modifier_function=None, 122 center_bias=True): 123 """Initializes a GradientBoostedDecisionTreeRegressor estimator instance. 124 125 Args: 126 learner_config: A config for the learner. 127 examples_per_layer: Number of examples to accumulate before growing a 128 layer. It can also be a function that computes the number of examples 129 based on the depth of the layer that's being built. 130 label_dimension: Number of regression labels per example. This is the size 131 of the last dimension of the labels `Tensor` (typically, this has shape 132 `[batch_size, label_dimension]`). 133 num_trees: An int, number of trees to build. 134 feature_columns: A list of feature columns. 135 label_name: String, name of the key in label dict. Can be null if label 136 is a tensor (single headed models). 137 weight_column_name: Name of the column for weights, or None if not 138 weighted. 139 model_dir: Directory for model exports, etc. 140 config: `RunConfig` object to configure the runtime settings. 141 feature_engineering_fn: Feature engineering function. Takes features and 142 labels which are the output of `input_fn` and returns features and 143 labels which will be fed into the model. 144 logits_modifier_function: A modifier function for the logits. 145 center_bias: Whether a separate tree should be created for first fitting 146 the bias. 147 """ 148 head = head_lib.regression_head( 149 label_name=label_name, 150 label_dimension=label_dimension, 151 weight_column_name=weight_column_name, 152 enable_centered_bias=False) 153 if label_dimension == 1: 154 learner_config.num_classes = 2 155 else: 156 learner_config.num_classes = label_dimension 157 super(GradientBoostedDecisionTreeRegressor, self).__init__( 158 model_fn=model.model_builder, 159 params={ 160 'head': head, 161 'feature_columns': feature_columns, 162 'learner_config': learner_config, 163 'num_trees': num_trees, 164 'weight_column_name': weight_column_name, 165 'examples_per_layer': examples_per_layer, 166 'logits_modifier_function': logits_modifier_function, 167 'center_bias': center_bias, 168 }, 169 model_dir=model_dir, 170 config=config, 171 feature_engineering_fn=feature_engineering_fn) 172 173 174 class GradientBoostedDecisionTreeEstimator(estimator.Estimator): 175 """An estimator using gradient boosted decision trees. 176 177 Useful for training with user specified `Head`. 178 """ 179 180 def __init__(self, 181 learner_config, 182 examples_per_layer, 183 head, 184 num_trees=None, 185 feature_columns=None, 186 weight_column_name=None, 187 model_dir=None, 188 config=None, 189 feature_engineering_fn=None, 190 logits_modifier_function=None, 191 center_bias=True): 192 """Initializes a GradientBoostedDecisionTreeEstimator estimator instance. 193 194 Args: 195 learner_config: A config for the learner. 196 examples_per_layer: Number of examples to accumulate before growing a 197 layer. It can also be a function that computes the number of examples 198 based on the depth of the layer that's being built. 199 head: `Head` instance. 200 num_trees: An int, number of trees to build. 201 feature_columns: A list of feature columns. 202 weight_column_name: Name of the column for weights, or None if not 203 weighted. 204 model_dir: Directory for model exports, etc. 205 config: `RunConfig` object to configure the runtime settings. 206 feature_engineering_fn: Feature engineering function. Takes features and 207 labels which are the output of `input_fn` and returns features and 208 labels which will be fed into the model. 209 logits_modifier_function: A modifier function for the logits. 210 center_bias: Whether a separate tree should be created for first fitting 211 the bias. 212 """ 213 super(GradientBoostedDecisionTreeEstimator, self).__init__( 214 model_fn=model.model_builder, 215 params={ 216 'head': head, 217 'feature_columns': feature_columns, 218 'learner_config': learner_config, 219 'num_trees': num_trees, 220 'weight_column_name': weight_column_name, 221 'examples_per_layer': examples_per_layer, 222 'logits_modifier_function': logits_modifier_function, 223 'center_bias': center_bias, 224 }, 225 model_dir=model_dir, 226 config=config, 227 feature_engineering_fn=feature_engineering_fn) 228
