1 # Copyright 2016 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 """Example of DNNClassifier for Iris plant dataset, with exponential decay.""" 15 16 from __future__ import absolute_import 17 from __future__ import division 18 from __future__ import print_function 19 20 import numpy as np 21 from sklearn import datasets 22 from sklearn import metrics 23 from sklearn import model_selection 24 import tensorflow as tf 25 26 27 X_FEATURE = 'x' # Name of the input feature. 28 29 30 def my_model(features, labels, mode): 31 """DNN with three hidden layers.""" 32 # Create three fully connected layers respectively of size 10, 20, and 10. 33 net = features[X_FEATURE] 34 for units in [10, 20, 10]: 35 net = tf.layers.dense(net, units=units, activation=tf.nn.relu) 36 37 # Compute logits (1 per class). 38 logits = tf.layers.dense(net, 3, activation=None) 39 40 # Compute predictions. 41 predicted_classes = tf.argmax(logits, 1) 42 if mode == tf.estimator.ModeKeys.PREDICT: 43 predictions = { 44 'class': predicted_classes, 45 'prob': tf.nn.softmax(logits) 46 } 47 return tf.estimator.EstimatorSpec(mode, predictions=predictions) 48 49 # Compute loss. 50 loss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits) 51 52 # Create training op with exponentially decaying learning rate. 53 if mode == tf.estimator.ModeKeys.TRAIN: 54 global_step = tf.train.get_global_step() 55 learning_rate = tf.train.exponential_decay( 56 learning_rate=0.1, global_step=global_step, 57 decay_steps=100, decay_rate=0.001) 58 optimizer = tf.train.AdagradOptimizer(learning_rate=learning_rate) 59 train_op = optimizer.minimize(loss, global_step=global_step) 60 return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op) 61 62 # Compute evaluation metrics. 63 eval_metric_ops = { 64 'accuracy': tf.metrics.accuracy( 65 labels=labels, predictions=predicted_classes) 66 } 67 return tf.estimator.EstimatorSpec( 68 mode, loss=loss, eval_metric_ops=eval_metric_ops) 69 70 71 def main(unused_argv): 72 iris = datasets.load_iris() 73 x_train, x_test, y_train, y_test = model_selection.train_test_split( 74 iris.data, iris.target, test_size=0.2, random_state=42) 75 76 classifier = tf.estimator.Estimator(model_fn=my_model) 77 78 # Train. 79 train_input_fn = tf.estimator.inputs.numpy_input_fn( 80 x={X_FEATURE: x_train}, y=y_train, num_epochs=None, shuffle=True) 81 classifier.train(input_fn=train_input_fn, steps=1000) 82 83 # Predict. 84 test_input_fn = tf.estimator.inputs.numpy_input_fn( 85 x={X_FEATURE: x_test}, y=y_test, num_epochs=1, shuffle=False) 86 predictions = classifier.predict(input_fn=test_input_fn) 87 y_predicted = np.array(list(p['class'] for p in predictions)) 88 y_predicted = y_predicted.reshape(np.array(y_test).shape) 89 90 # Score with sklearn. 91 score = metrics.accuracy_score(y_test, y_predicted) 92 print('Accuracy (sklearn): {0:f}'.format(score)) 93 94 # Score with tensorflow. 95 scores = classifier.evaluate(input_fn=test_input_fn) 96 print('Accuracy (tensorflow): {0:f}'.format(scores['accuracy'])) 97 98 99 if __name__ == '__main__': 100 tf.app.run() 101
