Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

関数プログラマから見たPythonと機械学習

1,277 views

Published on

2018-02-27に開催された 機械学習・ディープラーニングでPythonを使う嬉しさとツラミを分かちあう会 https://mlxse.connpass.

Published in: Engineering

関数プログラマから見たPythonと機械学習

  1. 1. Python (2018-02-27) Masahiro Sakai ( ) PYTHON
  2. 2. • Twitter: @masahiro_sakai github: https://github.com/msakai/ G+: https://plus.google.com/+MasahiroSakai • : ( ) • : – ( Haskell) – –
  3. 3. • • Python • Python • ……
  4. 4. • Haskell – Haskell ( ) • Ruby – ( Python ) – Ruby-GNOME2 • Python – Chainer + α – PyPI
  5. 5. DISCLAIMER • Python – – – • Python –
  6. 6. • • • •
  7. 7. • – R MATLAB • – numpy/scipy, pandas, scikit-learn, matplotlib, Jupyter Notebook, –
  8. 8. Python • …… Python • ) numpy.array == – numpy … – Python ⇒ – sorted([np.zeros(2), np.zeros(2)]) • ( collections-extended RangeMap )
  9. 9. : …… • …… • …… • ) pandas itertuples type(next(pd.DataFrame({"x" + str(i): [i] for i in range(253)}).itertuples())) #=> pandas.core.frame.Pandas type(next(pd.DataFrame({"x" + str(i): [i] for i in range(254)}).itertuples())) #=> tuple
  10. 10. : Lisp-1 • – Scheme, JavaScript – Common Lisp, Ruby Lisp-2 ( ) • Python Ruby obj.foo(args) obj.foo(args) ( ) obj.foo() obj.foo(), obj.foo obj.foo obj.method(:foo) m(args) m.call(args), m[args]
  11. 11. : Lisp-1 • …… → → • Ruby Lisp-2 • • Python Ruby obj.foo(…) obj.foo(…), obj.foo … ( ) obj.foo() obj.foo(), obj.foo obj.foo obj.method(:foo) m(…) m.call(…), m[…]
  12. 12. • (PEP 484) • mypy def greeting(name: str) -> str: return 'Hello ' + name print(greeting(0)) # => error: Argument 1 to "greeting" has incompatible type "int"; expected "str"
  13. 13. • Haskeller – – • – → • Python Haskell – • List Tuple List ……
  14. 14. • (gradual typing) – (dynamic typing) (static typing) – Jeremy G. Siek and Walid Taha, Gradual Typing for Functional Languages, in Proceedings of the 2006 Scheme and Functional Programming Workshop • – dynamic type: Any … Any – type consistency relation: A B • A = B Any ( : int→Any Any→str) – int ⊑ Any ⊑ str • (1)
  15. 15. (2) • – Python (mypy), TypeScript, Closure, Hack, Flow, C#? – • Python – • Python (mypy) • …… – (Any ) • e.g. Blame Calculus ( ) • Python (mypy)
  16. 16. • : • (np.array, chainer.Variable ) ⇒ • shape – shape • ( : forward backward ) (1) def __call__(self, x : np.array) -> np.array:
  17. 17. • shape Array(m,n,o) …… – reshape ⇒ (c.f. ) – Haskell hmatrix API † …… • (2) † https://hackage.haskell.org/package/hmatrix-0.18.2.0/docs/Numeric-LinearAlgebra-Static.html
  18. 18. : Chainer • Chainer DSL – – type_check.expect(in_types[0].shape == in_types[1].shape) Expect: in_types[0].shape == in_types[1].shape Actual: (3,) != (2,) http://blog.unnono.net/2015/10/chainer-typecheck.html
  19. 19. (3) • • … •
  20. 20. • CPython • GPU CPU – Python …… – NN CPU – GPU ……
  21. 21. numpy • numpy • numpy ( numpy ) – : x[np.arange(x.shape[0])[:, None], p] – •
  22. 22. • ( 1 540 2.1GB ) parsy-1.1.0 290m7.032s pyparsing-2.2.0 562m41.623s funcparserlib-0.3.6 79m45.315s megaparsec-5.3.1 (Haskell) 18m12.025s attoparsec-0.13.1.0 (Haskell) 15m46.415s : Python str Haskell ByteString(bytes )
  23. 23. • : – – …… • Cython – –
  24. 24. • – CPython (Go ) • GIL (Global Interpreter Lock) – CPython …… ONNX-Chainer ……
  25. 25. • Python • Python – – Python •

×
Save this presentationTap To Close