Prepare TorchFold for PyPI

Author: Maksym Zavershynskyi

README.md

@@ -1,13 +1,18 @@
-# pytorch-tools
-Tools for PyTorch
+# TorchFold
 
-## Torch Fold
+Blog post: http://near.ai/articles/2017-09-06-PyTorch-Dynamic-Batching/
 
 Analogous to [TensorFlow Fold](https://github.com/tensorflow/fold), implements dynamic batching with super simple interface.
 Replace every direct call in your computation to nn module with `f.add('function name', arguments)`.
-It will construct an optimized version of computation and on `f.apply` will dynmically batch and execute the computation on given nn module.
+It will construct an optimized version of computation and on `f.apply` will dynamically batch and execute the computation on given nn module.
 
-For example:
+## Installation
+We recommend using pip package manager:
+```
+pip install torchfold
+```
+
+## Example
 
 ```
     f = torchfold.Fold()
@@ -35,9 +40,3 @@ For example:
     model = Model(...)
     f.apply(model, [[res]])
 ```
-
-## Embeddings
-
-Many times you find yourself with new words in vocabulary as you are working on your model.
-Instead of re-training you can use `embeddings.expand_embeddings` to expand them on the fly with given vocabulary.
-

examples/snli/spinn-example.py

@@ -8,12 +8,11 @@
 from torch.autograd import Variable
 import torch.nn as nn
 from torch import optim
-import torch.nn.functional as F
 
 from torchtext import data
 from torchtext import datasets
 
-from pytorch_tools import torchfold
+import torchfold
 
 
 parser = argparse.ArgumentParser(description='SPINN')
@@ -97,7 +96,6 @@ def is_leaf(self):
         def __repr__(self):
             return str(self.id) if self.is_leaf() else "(%s, %s)" % (self.left, self.right)
 
-
     def __init__(self, example, inputs_vocab, answer_vocab):
         self.label = answer_vocab.stoi[example.label] - 1
         queue = []

pytorch_tools/__init__.py

@@ -1,13 +1,26 @@
-from setuptools import setup
+from setuptools import setup, find_packages
+from os import path
+
+VERSION = "0.1.0"
+
+here = path.abspath(path.dirname(__file__))
+with open(path.join(here, 'README.md'), encoding='utf-8') as f:
+    long_description = f.read()
 
-VERSION = "0.0.1"
 
 setup(
-    name='pytorch-tools',
+    name='torchfold',
     version=VERSION,
-    description='Tools for PyTorch',
-    packages=['pytorch_tools'],
+    description='Dynamic Batching with PyTorch',
+    long_description=long_description,
+    long_description_content_type='text/markdown',
+    packages=find_packages(exclude=["*_test.py"]),
     license='Apache License, Version 2.0',
-    author='Illia Polosukhin',
+    author='Illia Polosukhin, NEAR Inc',
+    author_email="illia@near.ai",
+    project_urls={
+        'Blog Post': "http://near.ai/articles/2017-09-06-PyTorch-Dynamic-Batching/",
+        'Source': "https://github.com/nearai/torchfold",
+    },
 )
 

pytorch_tools/embeddings.py

@@ -0,0 +1 @@
+from .torchfold import Fold, Unfold

pytorch_tools/trainer.py

@@ -2,9 +2,6 @@
 
 import torch
 from torch.autograd import Variable
-import torch.nn as nn
-from torch import optim
-import torch.nn.functional as F
 
 
 class Fold(object):
@@ -38,7 +35,6 @@ def __repr__(self):
             return "[%d:%d]%s" % (
                 self.step, self.index, self.op)
 
-
     class ComputedResult(object):
         def __init__(self, batch_size, batched_result):
             self.batch_size = batch_size
@@ -79,7 +75,6 @@ def get(self, index, split_idx=-1):
                     self.result[split_idx] = torch.chunk(self.result[split_idx], self.batch_size)
                 return self.result[split_idx][index]
 
-
     def __init__(self, volatile=False, cuda=False):
         self.steps = collections.defaultdict(
             lambda: collections.defaultdict(list))
@@ -96,7 +91,7 @@ def add(self, op, *args):
         """Add op to the fold."""
         self.total_nodes += 1
         if not all([isinstance(arg, (
-            Fold.Node, int, torch.tensor._TensorBase, Variable)) for arg in args]):
+                Fold.Node, int, torch.tensor._TensorBase, Variable)) for arg in args]):
             raise ValueError(
                 "All args should be Tensor, Variable, int or Node, got: %s" % str(args))
         if args not in self.cached_nodes[op]:
@@ -116,8 +111,7 @@ def _batch_args(self, arg_lists, values):
                            for arg_item in arg[1:])
 
                 if arg[0].batch:
-                    batched_arg = values[arg[0].step][arg[0]
-                                                      .op].try_get_batched(arg)
+                    batched_arg = values[arg[0].step][arg[0].op].try_get_batched(arg)
                     if batched_arg is not None:
                         res.append(batched_arg)
                     else:
@@ -142,8 +136,7 @@ def _batch_args(self, arg_lists, values):
                     if isinstance(arg_item, Fold.Node):
                         assert arg_item.batch
                         r.append(arg_item.get(values))
-                    elif isinstance(arg_item, (torch.tensor._TensorBase,
-                        Variable)):
+                    elif isinstance(arg_item, (torch.tensor._TensorBase, Variable)):
                         r.append(arg_item)
                     else:
                         raise ValueError(

setup.py

@@ -1,10 +1,6 @@
-import collections
-
 import torch
 from torch.autograd import Variable
 import torch.nn as nn
-from torch import optim
-import torch.nn.functional as F
 
 import torchfold
 
@@ -111,4 +107,4 @@ def test_rnn_optimized_chunking(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main()