further changes in anli dataset

data/anli.py

@@ -1,16 +1,34 @@
-import pandas as pd
+import os
+
 import jsonlines
+import pandas as pd
+
 from data import dataset
 
 
 class ANLIDataset(dataset.DatasetExperiment):
 
     def __init__(self, *args):
         super().__init__(*args)
-        with jsonlines.open('./corpora/anli_v0.1/R3/test.jsonl') as jsonl_reader:
-            for o in jsonl_reader:
-                print(o)
+        self.hypothesis = []
+        self.reason = []
+        self.label = []
+        with jsonlines.open(os.path.join(self.data_dir, 'train.jsonl')) as jsonl_reader:
+            for instance in jsonl_reader:
+                self.hypothesis.append(instance['hypothesis'])
+                self.reason.append(instance['reason'])
+                self.label.append(instance['label'])
 
+        dataset = pd.DataFrame(
+            list(
+                zip(
+                    self.hypothesis,
+                    self.reason,
+                    self.label,
+                )
+            ),
+            columns=['hypothesis', 'reason', 'label']
+        )
         num_instances = len(dataset)
         self.num_train = num_instances * (1 - self.dev_ratio - self.test_ratio)
         self.num_dev = num_instances * self.dev_ratio
@@ -24,28 +42,28 @@ def train_set(self):
         return self.train
 
     def train_set_pairs(self):
-        return self.train[['question1', 'question2']].as_matrix()
+        return self.train[['hypothesis', 'reason']].as_matrix()
 
     def train_labels(self):
-        return self.train['is_duplicate'].as_matrix()
+        return pd.get_dummies(self.train['label']).as_matrix()
 
     def dev_set(self):
         return self.dev
 
     def dev_set_pairs(self):
-        return self.dev[['question1', 'question2']].as_matrix()
+        return self.dev[['hypothesis', 'reason']].as_matrix()
 
     def dev_labels(self):
-        return self.dev['is_duplicate'].as_matrix()
+        return self.dev['label'].as_matrix()
 
     def test_set(self):
         return self.test
 
     def test_set_pairs(self):
-        return self.test[['question1', 'question2']].as_matrix()
+        return self.test[['hypothesis', 'reason']].as_matrix()
 
     def test_labels(self):
-        return self.test['is_duplicate'].as_matrix()
+        return self.test['label'].as_matrix()
 
     def _data_path(self):
-        return 'corpora/QQP/'
+        return 'corpora/ANLI/R3'

data/dataset.py

@@ -79,12 +79,14 @@ def pick_train_mini_batch(self):
         train_idxs = np.arange(len(self._train_labels))
         np.random.shuffle(train_idxs)
         train_idxs = train_idxs[:self.num_dev_instances()]
-        return self.train_sen1[train_idxs], self.train_sen2[train_idxs], self._train_labels[
-            train_idxs]
-
-    def __str__(self):
-        return 'Dataset properties:\n ' \
-               'Number of training instances: {}\n ' \
-               'Number of dev instances: {}\n ' \
-               'Number of test instances: {}\n' \
-            .format(len(self._train_labels), len(self._dev_labels), len(self._test_labels))
+        mini_train1 = self.train_sen1[train_idxs]
+        mini_train2 = self.train_sen2[train_idxs]
+        mini_labels = self._train_labels[train_idxs]
+        return mini_train1, mini_train2, mini_labels
+
+        def __str__(self):
+            return 'Dataset properties:\n ' \
+                   'Number of training instances: {}\n ' \
+                   'Number of dev instances: {}\n ' \
+                   'Number of test instances: {}\n' \
+                .format(len(self._train_labels), len(self._dev_labels), len(self._test_labels))

run.py

@@ -15,6 +15,8 @@
 from utils.model_saver import ModelSaver
 from utils.other_utils import timer, set_visible_gpu, init_config
 
+log = tf.logging.info
+
 
 def train(main_config, model_config, model_name, experiment_name, dataset_name):
     main_cfg = MainConfig(main_config)
@@ -38,8 +40,10 @@ def train(main_config, model_config, model_name, experiment_name, dataset_name):
     num_batches = dataset_helper.num_batches
     model = model(max_sentence_len, vocabulary_size, main_config, model_config)
     model_saver = ModelSaver(main_cfg.model_dir, experiment_name, main_cfg.checkpoints_to_keep)
-    config = tf.ConfigProto(allow_soft_placement=True,
-                            log_device_placement=main_cfg.log_device_placement)
+    config = tf.ConfigProto(
+        allow_soft_placement=True,
+        log_device_placement=main_cfg.log_device_placement,
+    )
 
     with tf.Session(config=config) as session:
         global_step = 0
@@ -50,6 +54,8 @@ def train(main_config, model_config, model_name, experiment_name, dataset_name):
 
         metrics = {'acc': 0.0}
         time_per_epoch = []
+
+        log('Training model for {} epochs'.format(main_cfg.num_epochs))
         for epoch in tqdm(range(main_cfg.num_epochs), desc='Epochs'):
             start_time = time.time()
 

utils/model_evaluator.py

@@ -17,10 +17,12 @@ def _evaluate(self, x1, x2, labels, batch_size=100):
         accuracy = 0.0
         for batch in range(num_batches):
             x1_batch, x2_batch, y_batch = batch_helper.next(batch)
-            feed_dict = {self._model.x1: x1_batch,
-                         self._model.x2: x2_batch,
-                         self._model.is_training: False,
-                         self._model.labels: y_batch}
+            feed_dict = {
+                self._model.x1: x1_batch,
+                self._model.x2: x2_batch,
+                self._model.is_training: False,
+                self._model.labels: y_batch
+            }
             accuracy += self._session.run(self._model.accuracy, feed_dict=feed_dict)
         accuracy /= num_batches
         return accuracy