PC-2016.3.2 <[email protected] Merge branch 'master'

README.md

@@ -21,6 +21,8 @@ optional arguments:
   -h, --help            show this help message and exit
   --embedding_dim EMBEDDING_DIM
                         Dimensionality of character embedding (default: 128)
+  --enable_word_embeddings
+                        Enable/disable the word embeddings (default: True)
   --filter_sizes FILTER_SIZES
                         Comma-separated filter sizes (default: '3,4,5')
   --num_filters NUM_FILTERS

config.yml

@@ -1,4 +1,6 @@
 word_embeddings:
+  # Two types of word embedding algorithm (word2vec and glove) are supported.
+  # Just set the default to empty string to disable the word embeddings
   default: word2vec
   word2vec:
     path: ../../data/GoogleNews-vectors-negative300.bin
@@ -10,6 +12,7 @@ word_embeddings:
     length: 400000
 
 datasets:
+  # Support currently 3 datasets: mrpolarity, 20newsgroup and localdata
   default: 20newsgroup
   mrpolarity:
     positive_data_file:
@@ -19,7 +22,7 @@ datasets:
       path: "data/rt-polaritydata/rt-polarity.neg"
       info: "Data source for the negative data"
   20newsgroup:
-    # The dataset includes following newsgroup: 
+    # The dataset includes following 20 newsgroups:
     # alt.atheism, comp.windows.x, rec.sport.hockey, soc.religion.christian
     # comp.graphics, misc.forsale, sci.crypt, talk.politics.guns
     # comp.os.ms-windows.misc, rec.autos, sci.electronics, talk.politics.mideast
@@ -32,5 +35,22 @@ datasets:
       - soc.religion.christian
     shuffle: True
     random_state: 42
-
+  localdata:
+    # Load text files with categories as subfolder names.
+    # Individual samples are assumed to be files stored
+    # a two levels folder structure such as the following:
+    # container_folder/
+    #   category_1_folder/
+    #     file_1.txt file_2.txt ... file_42.txt
+    #   category_2_folder/
+    #     file_43.txt file_44.txt ...
+    #
+    # As an example, a SentenceCorpus dataset from
+    # https://archive.ics.uci.edu/ml/datasets/Sentence+Classification
+    # has been used. The dataset includes following 3 domains:
+    # arxiv, jdm and plos
+    container_path: ../../data/SentenceCorpus
+    categories:
+    shuffle: True
+    random_state: 42
 

data_helpers.py

@@ -1,11 +1,8 @@
 import numpy as np
 import re
 from sklearn.datasets import fetch_20newsgroups
+from sklearn.datasets import load_files
 
-subset="train"
-categories=['alt.atheism', 'comp.graphics', 'sci.med', 'soc.religion.christian']
-shuffle=True
-random_state=42
 
 def clean_str(string):
     """
@@ -48,10 +45,13 @@ def batch_iter(data, batch_size, num_epochs, shuffle=True):
             yield shuffled_data[start_index:end_index]
 
 
-def get_datasets_20newsgroup(subset=subset, categories=categories, shuffle=shuffle, random_state=random_state):
+def get_datasets_20newsgroup(subset='train', categories=None, shuffle=True, random_state=42):
     """
     Retrieve data from 20 newsgroups
+    :param subset: train, test or all
     :param categories: List of newsgroup name
+    :param shuffle: shuffle the list or not
+    :param random_state: seed integer to shuffle the dataset
     :return: data and labels of the newsgroup
     """
     datasets = fetch_20newsgroups(subset=subset, categories=categories, shuffle=shuffle, random_state=random_state)
@@ -77,6 +77,23 @@ def get_datasets_mrpolarity(positive_data_file, negative_data_file):
     return datasets
 
 
+def get_datasets_localdata(container_path=None, categories=None, load_content=True,
+                       encoding='utf-8', shuffle=True, random_state=42):
+    """
+    Load text files with categories as subfolder names.
+    Individual samples are assumed to be files stored a two levels folder structure.
+    :param container_path: The path of the container
+    :param categories: List of classes to choose, all classes are chosen by default (if empty or omitted)
+    :param shuffle: shuffle the list or not
+    :param random_state: seed integer to shuffle the dataset
+    :return: data and labels of the dataset
+    """
+    datasets = load_files(container_path=container_path, categories=categories,
+                          load_content=load_content, shuffle=shuffle, encoding=encoding,
+                          random_state=random_state)
+    return datasets
+
+
 def load_data_labels(datasets):
     """
     Load data and labels

eval.py

@@ -54,11 +54,13 @@
     y_test = np.argmax(y_test, axis=1)
     print("Total number of test examples: {}".format(len(y_test)))
 else:
-    x_raw = ["The number of reported cases of gonorrhea in Colorado increased",
-             "I am in the market for a 24-bit graphics card for a PC"]
-    y_test = [2, 1]
-    # x_raw = ["a masterpiece four years in the making", "everything is off."]
-    # y_test = [1, 0]
+    if dataset_name == "mrpolarity":
+        x_raw = ["a masterpiece four years in the making", "everything is off."]
+        y_test = [1, 0]
+    else:
+        x_raw = ["The number of reported cases of gonorrhea in Colorado increased",
+                 "I am in the market for a 24-bit graphics card for a PC"]
+        y_test = [2, 1]
 
 # Map data into vocabulary
 vocab_path = os.path.join(FLAGS.checkpoint_dir, "..", "vocab")
@@ -104,7 +106,7 @@
     correct_predictions = float(sum(all_predictions == y_test))
     print("Total number of test examples: {}".format(len(y_test)))
     print("Accuracy: {:g}".format(correct_predictions/float(len(y_test))))
-    print(metrics.classification_report(y_test, all_predictions, target_names=datasets.target_names))
+    print(metrics.classification_report(y_test, all_predictions, target_names=datasets['target_names']))
     print(metrics.confusion_matrix(y_test, all_predictions))
 
 # Save the evaluation to a csv

text_cnn.py

@@ -54,7 +54,7 @@ def __init__(
 
         # Combine all the pooled features
         num_filters_total = num_filters * len(filter_sizes)
-        self.h_pool = tf.concat(3, pooled_outputs)
+        self.h_pool = tf.concat(pooled_outputs, 3)
         self.h_pool_flat = tf.reshape(self.h_pool, [-1, num_filters_total])
 
         # Add dropout
@@ -75,7 +75,7 @@ def __init__(
 
         # CalculateMean cross-entropy loss
         with tf.name_scope("loss"):
-            losses = tf.nn.softmax_cross_entropy_with_logits(self.scores, self.input_y)
+            losses = tf.nn.softmax_cross_entropy_with_logits(logits=self.scores, labels=self.input_y)
             self.loss = tf.reduce_mean(losses) + l2_reg_lambda * l2_loss
 
         # Accuracy

train.py

@@ -10,20 +10,15 @@
 from tensorflow.contrib import learn
 import yaml
 
-with open("config.yml", 'r') as ymlfile:
-    cfg = yaml.load(ymlfile)
-
-embedding_name = cfg['word_embeddings']['default']
-embedding_dimension = cfg['word_embeddings'][embedding_name]['dimension']
-dataset_name = cfg["datasets"]["default"]
-
 # Parameters
 # ==================================================
 
 # Data loading params
 tf.flags.DEFINE_float("dev_sample_percentage", .1, "Percentage of the training data to use for validation")
 
 # Model Hyperparameters
+tf.flags.DEFINE_boolean("enable_word_embeddings", True, "Enable/disable the word embedding (default: True)")
+tf.flags.DEFINE_integer("embedding_dim", 128, "Dimensionality of character embedding (default: 128)")
 tf.flags.DEFINE_string("filter_sizes", "3,4,5", "Comma-separated filter sizes (default: '3,4,5')")
 tf.flags.DEFINE_integer("num_filters", 128, "Number of filters per filter size (default: 128)")
 tf.flags.DEFINE_float("dropout_keep_prob", 0.5, "Dropout keep probability (default: 0.5)")
@@ -46,6 +41,15 @@
     print("{}={}".format(attr.upper(), value))
 print("")
 
+with open("config.yml", 'r') as ymlfile:
+    cfg = yaml.load(ymlfile)
+
+dataset_name = cfg["datasets"]["default"]
+if FLAGS.enable_word_embeddings and cfg['word_embeddings']['default'] is not None:
+    embedding_name = cfg['word_embeddings']['default']
+    embedding_dimension = cfg['word_embeddings'][embedding_name]['dimension']
+else:
+    embedding_dimension = FLAGS.embedding_dim
 
 # Data Preparation
 # ==================================================
@@ -61,6 +65,11 @@
                                                      categories=cfg["datasets"][dataset_name]["categories"],
                                                      shuffle=cfg["datasets"][dataset_name]["shuffle"],
                                                      random_state=cfg["datasets"][dataset_name]["random_state"])
+elif dataset_name == "localdata":
+    datasets = data_helpers.get_datasets_localdata(container_path=cfg["datasets"][dataset_name]["container_path"],
+                                                     categories=cfg["datasets"][dataset_name]["categories"],
+                                                     shuffle=cfg["datasets"][dataset_name]["shuffle"],
+                                                     random_state=cfg["datasets"][dataset_name]["random_state"])
 x_text, y = data_helpers.load_data_labels(datasets)
 
 # Build vocabulary
@@ -148,23 +157,24 @@
 
         # Initialize all variables
         sess.run(tf.global_variables_initializer())
-        vocabulary = vocab_processor.vocabulary_
-        initW = None
-        if embedding_name == 'word2vec':
-            # load embedding vectors from the word2vec
-            print("Load word2vec file {}".format(cfg['word_embeddings']['word2vec']['path']))
-            initW = data_helpers.load_embedding_vectors_word2vec(vocabulary,
-                                                                 cfg['word_embeddings']['word2vec']['path'],
-                                                                 cfg['word_embeddings']['word2vec']['binary'])
-            print("word2vec file has been loaded")
-        elif embedding_name == 'glove':
-            # load embedding vectors from the glove
-            print("Load glove file {}".format(cfg['word_embeddings']['glove']['path']))
-            initW = data_helpers.load_embedding_vectors_glove(vocabulary,
-                                                              cfg['word_embeddings']['glove']['path'],
-                                                              embedding_dimension)
-            print("glove file has been loaded\n")
-        sess.run(cnn.W.assign(initW))
+        if FLAGS.enable_word_embeddings and cfg['word_embeddings']['default'] is not None:
+            vocabulary = vocab_processor.vocabulary_
+            initW = None
+            if embedding_name == 'word2vec':
+                # load embedding vectors from the word2vec
+                print("Load word2vec file {}".format(cfg['word_embeddings']['word2vec']['path']))
+                initW = data_helpers.load_embedding_vectors_word2vec(vocabulary,
+                                                                     cfg['word_embeddings']['word2vec']['path'],
+                                                                     cfg['word_embeddings']['word2vec']['binary'])
+                print("word2vec file has been loaded")
+            elif embedding_name == 'glove':
+                # load embedding vectors from the glove
+                print("Load glove file {}".format(cfg['word_embeddings']['glove']['path']))
+                initW = data_helpers.load_embedding_vectors_glove(vocabulary,
+                                                                  cfg['word_embeddings']['glove']['path'],
+                                                                  embedding_dimension)
+                print("glove file has been loaded\n")
+            sess.run(cnn.W.assign(initW))
 
         def train_step(x_batch, y_batch):
             """