Merge pull request #68 from PetrochukM/update

Release 0.4.0 - Encoder rewrite, variable sequence collate support, reduced memory usage, doctests, removed SRU

Author: PetrochukM

.style.yapf

@@ -0,0 +1,4 @@
+[style]
+based_on_style = chromium
+indent_width = 4
+column_limit = 100

.travis.yml

@@ -2,15 +2,20 @@ dist: trusty
 sudo: required
 
 language: python
-python:
-  - '3.6'
-  - '3.5'
+matrix:
+  include:
+    - python: 3.6
+      dist: trusty
+      sudo: false
+    - python: 3.7
+      dist: xenial
+      sudo: true
 
 cache: pip
 
 notifications:
   email: false
-  
+
 before_install: source build_tools/travis/before_install.sh
 install: source build_tools/travis/install.sh
 script: RUN_DOCS=true RUN_SLOW=true RUN_FLAKE8=true bash build_tools/travis/test_script.sh

README.md

@@ -19,7 +19,7 @@ Join our community, add datasets and neural network layers! Chat with us on [Git
 
 ## Installation
 
-Make sure you have Python 3.5+ and PyTorch 0.4 or newer. You can then install `pytorch-nlp` using
+Make sure you have Python 3.6+ and PyTorch 1.0+. You can then install `pytorch-nlp` using
 pip:
 
     pip install pytorch-nlp
@@ -50,35 +50,32 @@ train[0]  # RETURNS: {'text': 'For a movie that gets..', 'sentiment': 'pos'}
 
 ### Apply [Neural Networks](http://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.nn.html) Layers
 
-For example, from the neural network package, apply a Simple Recurrent Unit (SRU):
+For example, from the neural network package, apply state-of-the-art LockedDropout:
 
 ```python
-from torchnlp.nn import SRU
 import torch
+from torchnlp.nn import LockedDropout
 
-input_ = torch.autograd.Variable(torch.randn(6, 3, 10))
-sru = SRU(10, 20)
+input_ = torch.randn(6, 3, 10)
+dropout = LockedDropout(0.5)
 
-# Apply a Simple Recurrent Unit to `input_`
-sru(input_)
-# RETURNS: (
-#   output [torch.FloatTensor (6x3x20)],
-#   hidden_state [torch.FloatTensor (2x3x20)]
-# )
+# Apply a LockedDropout to `input_`
+dropout(input_)
+# RETURNS: torch.FloatTensor (6x3x10)
 ```
 
-### [Encode Text](http://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.text_encoders.html)
+### [Encode Text](http://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.encoders.text.html)
 
 Tokenize and encode text as a tensor. For example, a `WhitespaceEncoder` breaks text into terms whenever it encounters a whitespace character.
 
 ```python
-from torchnlp.text_encoders import WhitespaceEncoder
+from torchnlp.encoders.text import WhitespaceEncoder
 
 # Create a `WhitespaceEncoder` with a corpus of text
 encoder = WhitespaceEncoder(["now this ain't funny", "so don't you dare laugh"])
 
 # Encode and decode phrases
-encoder.encode("this ain't funny.") # RETURNS: torch.LongTensor([6, 7, 1])
+encoder.encode("this ain't funny.") # RETURNS: torch.Tensor([6, 7, 1])
 encoder.decode(encoder.encode("This ain't funny.")) # RETURNS: "this ain't funny."
 ```
 

build_tools/travis/install.sh

@@ -35,10 +35,10 @@ python -m spacy download en
 python -m nltk.downloader perluniprops nonbreaking_prefixes
 
 # Install PyTorch Dependancies
-if [[ $TRAVIS_PYTHON_VERSION == '3.6' ]]; then
-    pip install http://download.pytorch.org/whl/cpu/torch-0.4.0-cp36-cp36m-linux_x86_64.whl
+if [[ $TRAVIS_PYTHON_VERSION == '3.7' ]]; then
+    pip install https://download.pytorch.org/whl/cpu/torch-1.0.1.post2-cp37-cp37m-linux_x86_64.whl
 fi
-if [[ $TRAVIS_PYTHON_VERSION == '3.5' ]]; then
-    pip install http://download.pytorch.org/whl/cpu/torch-0.4.0-cp35-cp35m-linux_x86_64.whl
+if [[ $TRAVIS_PYTHON_VERSION == '3.6' ]]; then
+    pip install https://download.pytorch.org/whl/cpu/torch-1.0.1.post2-cp36-cp36m-linux_x86_64.whl
 fi
 pip install torchvision

build_tools/travis/test_script.sh

@@ -7,22 +7,23 @@
 # Exit immediately if a command exits with a non-zero status.
 set -e
 
-python --version
-
-if [[ "$RUN_FLAKE8" == "true" ]]; then
-    flake8
-fi
+export PYTHONPATH=.
 
+python --version
 
 if [[ "$RUN_DOCS" == "true" ]]; then
     make -C docs html
 fi
 
+if [[ "$RUN_FLAKE8" == "true" ]]; then
+    flake8 torchnlp/
+    flake8 tests/
+fi
+
 run_tests() {
+    TEST_CMD="python -m pytest tests/ torchnlp/ --verbose --durations=20 --cov=torchnlp --doctest-modules"
     if [[ "$RUN_SLOW" == "true" ]]; then
-        TEST_CMD="py.test -v --durations=20 --cov=torchnlp --runslow"
-    else
-        TEST_CMD="py.test -v --durations=20 --cov=torchnlp"
+        TEST_CMD="$TEST_CMD --runslow"
     fi
     $TEST_CMD
 }

docs/index.rst

@@ -16,7 +16,7 @@ and text encoders. It's open-source software, released under the BSD3 license.
    source/torchnlp.datasets
    source/torchnlp.word_to_vector
    source/torchnlp.nn
-   source/torchnlp.text_encoders
+   source/torchnlp.encoders
    source/torchnlp.samplers
    source/torchnlp.metrics
    source/torchnlp.utils

docs/source/torchnlp.encoders.rst

@@ -0,0 +1,15 @@
+torchnlp.encoders package
+===============================
+
+The ``torchnlp.encoders`` package supports encoding objects as a vector
+:class:`torch.Tensor` and decoding a vector :class:`torch.Tensor` back.
+
+.. automodule:: torchnlp.encoders
+    :members:
+    :undoc-members:
+    :show-inheritance:
+
+.. automodule:: torchnlp.encoders.text
+    :members:
+    :undoc-members:
+    :show-inheritance:

docs/source/torchnlp.text_encoders.rst

@@ -1,4 +1,4 @@
-`awd-lstm-lm` set the state-of-the-art in word level perplexities in 2017. With PyTorch NLP, we show that in 30 minutes, we were able to reduce the footprint of this repository by 4 files (185 lines of code). We employ the use of the [datasets package](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html), [IdentityEncoder module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.text_encoders.html#torchnlp.text_encoders.IdentityEncoder), [BPTTBatchSampler module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.samplers.html#torchnlp.samplers.BPTTBatchSampler), [LockedDropout module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.nn.html#torchnlp.nn.LockedDropout) and [WeightDrop module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.nn.html#torchnlp.nn.WeightDrop)
+`awd-lstm-lm` set the state-of-the-art in word level perplexities in 2017. With PyTorch NLP, we show that in 30 minutes, we were able to reduce the footprint of this repository by 4 files (185 lines of code). We employ the use of the [datasets package](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.datasets.html), [IdentityEncoder module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.encoders.text.html#torchnlp.encoders.text.IdentityEncoder), [BPTTBatchSampler module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.samplers.html#torchnlp.samplers.BPTTBatchSampler), [LockedDropout module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.nn.html#torchnlp.nn.LockedDropout) and [WeightDrop module](https://pytorchnlp.readthedocs.io/en/latest/source/torchnlp.nn.html#torchnlp.nn.WeightDrop)
 
 
 Below is the original README from the repository:

examples/awd-lstm-lm/README.md

@@ -96,17 +96,17 @@ def model_load(fn):
 
 
 from torchnlp import datasets
-from torchnlp.text_encoders import IdentityEncoder
+from torchnlp.encoders import LabelEncoder
 from torchnlp.samplers import BPTTBatchSampler
 
 print('Producing dataset...')
 train, val, test = getattr(datasets, args.data)(train=True, dev=True, test=True)
 
-encoder = IdentityEncoder(train + val + test)
+encoder = LabelEncoder(train + val + test)
 
-train_data = encoder.encode(train)
-val_data = encoder.encode(val)
-test_data = encoder.encode(test)
+train_data = encoder.batch_encode(train)
+val_data = encoder.batch_encode(val)
+test_data = encoder.batch_encode(test)
 
 eval_batch_size = 10
 test_batch_size = 1