PyTorch 1.3 + TorchText 0.4 fixes

.travis.yml

@@ -1,26 +1,19 @@
 matrix:
   include:
-  # - os: linux
-  #   python: 2.7
-  #   env: PYTHON_VERSION=2.7
-  - os: linux
-    python: 3.5
-    env: PYTHON_VERSION=3.5
   - os: linux
     python: 3.6
     env: PYTHON_VERSION=3.6
-  # - os: osx
-  #   language: generic
-  #   env:
-  #   - PYTHON_VERSION=2.7
+  - os: linux
+    python: 3.7
+    env: PYTHON_VERSION=3.7
   - os: osx
     language: generic
     env:
-    - PYTHON_VERSION=3.5
+    - PYTHON_VERSION=3.6
   - os: osx
     language: generic
     env:
-    - PYTHON_VERSION=3.6
+    - PYTHON_VERSION=3.7
 
 notifications:
   email: false
@@ -39,7 +32,7 @@ before_install:
 
 install:
   - conda install --yes python=$PYTHON_VERSION pip scikit-learn nose
-  - pip install --process-dependency-links git+https://github.com/anhaidgroup/deepmatcher | cat
+  - pip install -e . | cat
   - python -m nltk.downloader perluniprops nonbreaking_prefixes punkt
 
 script:

deepmatcher/batch.py

@@ -48,7 +48,7 @@ def __new__(cls, *args, **kwargs):
             if 'word_probs' in train_info.metadata:
                 raw_word_probs = train_info.metadata['word_probs'][name]
                 word_probs = torch.Tensor(
-                    [[raw_word_probs[w] for w in b] for b in data.data])
+                    [[raw_word_probs[int(w)] for w in b] for b in data.data])
                 if data.is_cuda:
                     word_probs = word_probs.cuda()
             pc = None

deepmatcher/data/dataset.py

@@ -225,15 +225,15 @@ def compute_metadata(self, pca=False):
 
         # Create an iterator over the entire dataset.
         train_iter = MatchingIterator(
-            self, self, train=False, batch_size=1024, device=-1, sort_in_buckets=False)
+            self, self, train=False, batch_size=1024, device='cpu', sort_in_buckets=False)
         counter = defaultdict(Counter)
 
         # For each attribute, find the number of times each word id occurs in the dataset.
         # Note that word ids here also include ``UNK`` tokens, padding tokens, etc.
         for batch in pyprind.prog_bar(train_iter, title='\nBuilding vocabulary'):
             for name in self.all_text_fields:
                 attr_input = getattr(batch, name)
-                counter[name].update(attr_input.data.data.view(-1))
+                counter[name].update(attr_input.data.data.view(-1).tolist())
 
         word_probs = {}
         totals = {}
@@ -270,7 +270,7 @@ def compute_metadata(self, pca=False):
 
         # Create an iterator over the entire dataset.
         train_iter = MatchingIterator(
-            self, self, train=False, batch_size=1024, device=-1, sort_in_buckets=False)
+            self, self, train=False, batch_size=1024, device='cpu', sort_in_buckets=False)
         attr_embeddings = defaultdict(list)
 
         # Run the constructed neural network to compute weighted sequence embeddings

deepmatcher/data/field.py

@@ -10,6 +10,7 @@
 import torch
 from torchtext import data, vocab
 from torchtext.utils import download_from_url
+from urllib.request import urlretrieve
 
 logger = logging.getLogger(__name__)
 
@@ -23,6 +24,7 @@ def __init__(self,
         url = url_base + suffix
         base, ext = os.path.splitext(suffix)
         name = suffix if ext == '.vec' else base
+        print('Fasttext url b', url_base)
         super(FastText, self).__init__(name, url=url, **kwargs)
 
 
@@ -60,7 +62,10 @@ def cache(self, name, cache, url=None):
             if not os.path.exists(cache):
                 os.makedirs(cache)
             if not os.path.isfile(self.destination):
-                download_from_url(url, self.destination)
+                if 'drive.google.com' in url:
+                    download_from_url(url, self.destination)
+                else:
+                    urlretrieve(url, self.destination)
             logger.info('Extracting vectors into {}'.format(cache))
             ext = os.path.splitext(self.destination)[1][1:]
             if ext == 'zip':

deepmatcher/models/core.py

@@ -351,7 +351,7 @@ def initialize(self, train_dataset, init_batch=None):
                 train_dataset,
                 train=False,
                 batch_size=4,
-                device=-1,
+                device='cpu',
                 sort_in_buckets=False)
             init_batch = next(run_iter.__iter__())
         self.forward(init_batch)

deepmatcher/models/modules.py

@@ -721,13 +721,13 @@ def _forward(self, input_with_meta):
             if input_with_meta.lengths is not None:
                 mask = _utils.sequence_mask(input_with_meta.lengths)
                 mask = mask.unsqueeze(2)  # Make it broadcastable.
-                input.data.masked_fill_(1 - mask, -float('inf'))
+                input.data.masked_fill_(~mask, -float('inf'))
             output = input.max(dim=1)[0]
         else:
             if input_with_meta.lengths is not None:
                 mask = _utils.sequence_mask(input_with_meta.lengths)
                 mask = mask.unsqueeze(2)  # Make it broadcastable.
-                input.data.masked_fill_(1 - mask, 0)
+                input.data.masked_fill_(~mask, 0)
 
             lengths = Variable(input_with_meta.lengths.clamp(min=1).unsqueeze(1).float())
             if self.style == 'avg':
@@ -860,7 +860,7 @@ def _forward(self, transformed, raw):
                 res *= math.sqrt(0.5)
             return res
         elif self.style == 'highway':
-            transform_gate = F.sigmoid(self.highway_gate(raw) + self.highway_bias)
+            transform_gate = torch.sigmoid(self.highway_gate(raw) + self.highway_bias)
             carry_gate = 1 - transform_gate
             return transform_gate * transformed + carry_gate * adjusted_raw
 

deepmatcher/models/word_aggregators.py

@@ -137,7 +137,7 @@ def _forward(self, input_with_meta, context_with_meta):
 
         if input_with_meta.lengths is not None:
             mask = _utils.sequence_mask(input_with_meta.lengths)
-            alignment_scores.data.masked_fill_(1 - mask, -float('inf'))
+            alignment_scores.data.masked_fill_(~mask, -float('inf'))
 
         # Make values along dim 2 sum to 1.
         normalized_scores = self.softmax(alignment_scores)

deepmatcher/models/word_comparators.py

@@ -172,7 +172,7 @@ def _forward(self,
             if context_with_meta.lengths is not None:
                 mask = _utils.sequence_mask(context_with_meta.lengths)
                 mask = mask.unsqueeze(1)  # Make it broadcastable.
-                alignment_scores.data.masked_fill_(1 - mask, -float('inf'))
+                alignment_scores.data.masked_fill_(~mask, -float('inf'))
 
             # Make values along dim 2 sum to 1.
             normalized_scores = self.softmax(alignment_scores)

deepmatcher/models/word_contextualizers.py

@@ -156,7 +156,7 @@ def _forward(self, input_with_meta):
             if input_with_meta.lengths is not None:
                 mask = _utils.sequence_mask(input_with_meta.lengths)
                 mask = mask.unsqueeze(1)  # Make it broadcastable.
-                alignment_scores.data.masked_fill_(1 - mask, -float('inf'))
+                alignment_scores.data.masked_fill_(~mask, -float('inf'))
 
             normalized_scores = self.softmax(alignment_scores)
 

deepmatcher/optim.py

@@ -4,7 +4,7 @@
 import torch.nn as nn
 import torch.optim as optim
 from torch.autograd import Variable
-from torch.nn.utils import clip_grad_norm
+from torch.nn.utils import clip_grad_norm_
 
 logger = logging.getLogger('deepmatcher.optim')
 
@@ -143,7 +143,7 @@ def step(self):
         self._step += 1
 
         if self.max_grad_norm:
-            clip_grad_norm(self.params, self.max_grad_norm)
+            clip_grad_norm_(self.params, self.max_grad_norm)
         self.base_optimizer.step()
 
     def update_learning_rate(self, acc, epoch):

setup.py

@@ -37,6 +37,6 @@ def find_version(*file_paths):
     packages=['deepmatcher', 'deepmatcher.data', 'deepmatcher.models'],
     python_requires='>=3.5',
     install_requires=[
-        'torch==0.3.1', 'tqdm', 'pyprind', 'six', 'Cython', 'torchtext', 'nltk>=3.2.5',
+        'torch>=1.0', 'tqdm', 'pyprind', 'six', 'Cython', 'torchtext', 'nltk>=3.2.5',
         'fasttextmirror', 'pandas'
     ])

test/test_dataset.py

@@ -12,13 +12,8 @@
 from deepmatcher.data.field import FastText, MatchingField
 from deepmatcher.data.process import _make_fields, process
 from torchtext.utils import unicode_csv_reader
-
-try:
-    from urllib.parse import urljoin
-    from urllib.request import pathname2url
-except ImportError:
-    from urlparse import urljoin
-    from urllib import path2pathname2url
+from urllib.parse import urljoin
+from urllib.request import pathname2url
 
 
 # import nltk

test/test_field.py

@@ -11,13 +11,8 @@
 from deepmatcher.data.field import (FastText, FastTextBinary, MatchingField,
                                     MatchingVocab, reset_vector_cache)
 from torchtext.vocab import Vectors
-
-try:
-    from urllib.parse import urljoin
-    from urllib.request import pathname2url
-except ImportError:
-    from urlparse import urljoin
-    from urllib import path2pathname2url
+from urllib.parse import urljoin
+from urllib.request import pathname2url
 
 # import nltk
 # nltk.download('perluniprops')
@@ -226,10 +221,10 @@ def test_extend_vectors_1(self):
         v.unk_init = torch.Tensor.zero_
         tokens = {'hello', 'world'}
         v.extend_vectors(tokens, vec_data)
-        self.assertEqual(len(v.itos), 3)
-        self.assertEqual(v.vectors.size(), torch.Size([3, 300]))
-        self.assertEqual(list(v.vectors[1][0:10]), [0.0] * 10)
+        self.assertEqual(len(v.itos), 4)
+        self.assertEqual(v.vectors.size(), torch.Size([4, 300]))
         self.assertEqual(list(v.vectors[2][0:10]), [0.0] * 10)
+        self.assertEqual(list(v.vectors[3][0:10]), [0.0] * 10)
 
         if os.path.exists(vectors_cache_dir):
             shutil.rmtree(vectors_cache_dir)

test/test_integration.py

@@ -11,13 +11,8 @@
 from deepmatcher.data.field import MatchingField, FastText
 from deepmatcher.data.process import process, process_unlabeled
 from deepmatcher import MatchingModel
-
-try:
-    from urllib.parse import urljoin
-    from urllib.request import pathname2url
-except ImportError:
-    from urlparse import urljoin
-    from urllib import path2pathname2url
+from urllib.parse import urljoin
+from urllib.request import pathname2url
 
 from test import test_dir_path
 

test/test_iterator.py

@@ -9,13 +9,8 @@
 from deepmatcher.data.iterator import MatchingIterator
 
 from test import test_dir_path
-
-try:
-    from urllib.parse import urljoin
-    from urllib.request import pathname2url
-except ImportError:
-    from urlparse import urljoin
-    from urllib import path2pathname2url
+from urllib.parse import urljoin
+from urllib.request import pathname2url
 
 class ClassMatchingIteratorTestCases(unittest.TestCase):
     def test_splits_1(self):

test/test_process.py

@@ -10,13 +10,8 @@
 from deepmatcher.data.process import _check_header, _make_fields, process, process_unlabeled
 from torchtext.utils import unicode_csv_reader
 from deepmatcher import MatchingModel
-
-try:
-    from urllib.parse import urljoin
-    from urllib.request import pathname2url
-except ImportError:
-    from urlparse import urljoin
-    from urllib import path2pathname2url
+from urllib.parse import urljoin
+from urllib.request import pathname2url
 
 
 class CheckHeaderTestCases(unittest.TestCase):