RNN文本分类、预测与服务

Question

我试图构建预测下一个单词（在我的例子中是URL）的模型。按照我的例子，我被困在预测部分。 我的python代码：

import argparse import sys import os import re import numpy as np import pandas import tensorflow as tf import url_datasets from tensorflow.contrib.learn.python.learn.preprocessing import text from tensorflow.python.framework import dtypes tf.app.flags.DEFINE_integer('model_version', 1, 'version number of the model.') tf.app.flags.DEFINE_string('work_dir', '/tmp/suc', 'Working directory.') FLAGS = tf.app.flags.FLAGS MAX_DOCUMENT_LENGTH = 40 EMBEDDING_SIZE = 40 n_words = 0 MAX_LABEL = 50 WORDS_FEATURE = 'words' # Name of the input words feature. TOKENIZER_RE = re.compile(r'([/a-z_-]*)\s') def tokenizer(iterator): """Tokenizer generator. Args: iterator: Input iterator with strings. Yields: array of tokens per each value in the input. """ for value in iterator: print(value) print(TOKENIZER_RE.findall(value)) yield TOKENIZER_RE.findall(value) def estimator_spec_for_softmax_classification( logits, labels, mode): """Returns EstimatorSpec instance for softmax classification.""" predicted_classes = tf.argmax(logits, 1) if mode == tf.estimator.ModeKeys.PREDICT: return tf.estimator.EstimatorSpec( mode=mode, predictions={ 'class': predicted_classes, 'prob': tf.nn.softmax(logits) }) onehot_labels = tf.one_hot(labels, MAX_LABEL, 1, 0) loss = tf.losses.softmax_cross_entropy( onehot_labels=onehot_labels, logits=logits) if mode == tf.estimator.ModeKeys.TRAIN: optimizer = tf.train.AdamOptimizer(learning_rate=0.1) train_op = optimizer.minimize(loss, global_step=tf.train.get_global_step()) return tf.estimator.EstimatorSpec(mode, loss=loss, train_op=train_op) eval_metric_ops = { 'accuracy': tf.metrics.accuracy( labels=labels, predictions=predicted_classes) } return tf.estimator.EstimatorSpec( mode=mode, loss=loss, eval_metric_ops=eval_metric_ops) def updatePrediction(prediction): file = open("/tmp/ai/prediction.txt","a") file.write(str(prediction)) file.close() def rnn_model(features, labels, mode): word_vectors = tf.contrib.layers.embed_sequence( features[WORDS_FEATURE], vocab_size=n_words, embed_dim=EMBEDDING_SIZE) word_list = tf.unstack(word_vectors, axis=1) cell = tf.contrib.rnn.GRUCell(EMBEDDING_SIZE) _, encoding = tf.contrib.rnn.static_rnn(cell, word_list, dtype=tf.float32) logits = tf.layers.dense(encoding, MAX_LABEL, activation=None) return estimator_spec_for_softmax_classification( logits=logits, labels=labels, mode=mode) def main(_): sess = tf.InteractiveSession() serialized_tf_example = tf.placeholder(tf.string, name='tf_example') global n_words urls = url_datasets.load_urls('/tmp/ai/demo') x_train = pandas.Series(urls.train.data[:,1]) labels = pandas.Series(urls.train.data[:,0]) y_train = pandas.Series(urls.train.target) x_test = pandas.Series(urls.test.data[:,1]) y_test = pandas.Series(urls.test.target) vocab_processor = text.VocabularyProcessor(MAX_DOCUMENT_LENGTH, min_frequency=0, tokenizer_fn=tokenizer) vocab_processor.fit(labels) x_transform_train = vocab_processor.fit_transform(x_train) x_transform_test = vocab_processor.transform(x_test) x_train = np.array(list(x_transform_train)) x_test = np.array(list(x_transform_test)) print(vocab_processor.vocabulary_._mapping) n_words = len(vocab_processor.vocabulary_) vocab_dict = vocab_processor.vocabulary_._mapping model_fn = rnn_model classifier = tf.estimator.Estimator(model_fn=model_fn) train_input_fn = tf.estimator.inputs.numpy_input_fn( x={WORDS_FEATURE: x_train}, y=y_train, batch_size=len(x_train), num_epochs=None, shuffle=False) classifier.train(input_fn=train_input_fn, steps=100) test_input_fn = tf.estimator.inputs.numpy_input_fn( x={WORDS_FEATURE: x_test}, y=y_test, num_epochs=1, shuffle=False) predictions = classifier.predict(input_fn=test_input_fn) export_path_base = sys.argv[-1] export_path = os.path.join( tf.compat.as_bytes(export_path_base), tf.compat.as_bytes(str(FLAGS.model_version))) print('Exporting trained model to', export_path) builder = tf.saved_model.builder.SavedModelBuilder(export_path) y_predicted = np.array(list(p['class'] for p in predictions)) inverseDictionary = dict(zip(vocab_dict.values(), vocab_dict.keys())) for prediction in y_predicted: print("prediction:"+inverseDictionary[int(prediction)]) updatePrediction(prediction) print("--> %s" % prediction) tensor_info_x = tf.saved_model.utils.build_tensor_info(serialized_tf_example) tensor_info_y = tf.saved_model.utils.build_tensor_info(tf.convert_to_tensor(y_predicted, tf.float32)) classification_inputs = tf.saved_model.utils.build_tensor_info( serialized_tf_example) classification_signature = ( tf.saved_model.signature_def_utils.build_signature_def( inputs={ tf.saved_model.signature_constants.CLASSIFY_INPUTS: classification_inputs }, outputs={ tf.saved_model.signature_constants.CLASSIFY_OUTPUT_SCORES: tensor_info_y }, method_name=tf.saved_model.signature_constants.CLASSIFY_METHOD_NAME)) prediction_signature = ( tf.saved_model.signature_def_utils.build_signature_def( inputs={'x_strings': tensor_info_x}, outputs={'scores': tensor_info_y}, method_name=tf.saved_model.signature_constants.PREDICT_METHOD_NAME)) legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op') builder.add_meta_graph_and_variables( sess, [tf.saved_model.tag_constants.SERVING], signature_def_map={ 'predict_url': prediction_signature, tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY: classification_signature, }, legacy_init_op=legacy_init_op) builder.save() print('Done exporting!') if __name__ == '__main__': parser = argparse.ArgumentParser() parser.add_argument( '--test_with_fake_data', default=False, help='Test the example code with fake data.', action='store_true') parser.add_argument( '--bow_model', default=False, help='Run with BOW model instead of RNN.', action='store_true') tf.app.run()