Contents
- Type Annotation
- Data Classes
- Packaging and Dependency Management
Type Annotation 
Some benefits of type annotations
- Static code analysis prior to runtime
- Self-documenting
- Code itself is a document
- Code completion
train, val, test = chainer.datasets.get_ptb_words()
- Type of
train, val, test?
- What kind of operations can be done on these variables?
🤔
Go to Declaration
def get_ptb_words():
"""Gets the Penn Tree Bank dataset as long word sequences.
[...]
Returns:
tuple of numpy.ndarray: Int32 vectors of word IDs.
"""
train = _retrieve_ptb_words('train.npz', _train_url)
valid = _retrieve_ptb_words('valid.npz', _valid_url)
test = _retrieve_ptb_words('test.npz', _test_url)
return train, valid, test
- 😸 docstring
- 😿 IDEs can’t parse complicated types
Add Type Annotations
from typing import Dict
def load_vocab(vocabfile: str) -> Dict[int, str]:
[...]
vocab = load_vocab(vocabfile)
If you can’t modify function declaration…
def load_vocab(vocabfile):
[...]
from typing import Dict
vocab: Dict[int, str] = load_vocab(vocabfile)
🤗
Data Classes 
Easier Way to Write Classes 🤗
from dataclasses import dataclass
from typing import Dict, Optional
@dataclass
class Vocabulary:
padding_token: str
oov_token: str
max_vocab_size: int
pretrained_files: Optional[Dict[str, str]] = None
- Similar to Scala’s Case Classes
Conventional Class 🤔
class Vocabulary:
def __init__(
self,
padding_token: str,
oov_token: str,
max_vocab_size: int,
pretrained_files: Optional[Dict[str, str]] = None,
):
self.padding_token = padding_token
self.oov_token = oov_token
self.max_vocab_size = max_vocab_size
self.pretrained_files = pretrained_files
def __eq__(self, other):
if isinstance(self, other.__class__):
return self.__dict__ == other.__dict__
return False
- Documentation (en / ja)
- attrs library is also recommended
- more features than Data Classes
- move faster
Packaging and Dependency Management
- How to reproduce your environment?
- How to make your program
pip installable?
Other languages have great tools
- Scala:
sbt
- JavaScript:
npm, yarn
- Rust:
cargo
- Go:
dep
- Python: …🤔
Ancienty Way 🤔
- Dump all installed packages and force to download them all:
$ pip freeze > requirements.txt
$ pip install -r requirements.txt
- Need to manage virtual environment at your own responsibility
- Don’t you have unnecessary dependency?
- How is version compatibility of Python and libraries?
- Metadata management?
Medieval Way 🤔
- Use standard Setuptools
- Messy documentation
- Need to write messy
setup.py
Modern Way 🤗
$ poetry init
$ poetry add chainer
$ poetry build
$ poetry publish
Further Readings
- Type Annotation
- Data Classes
- Packaging and Dependency Management