spaCy v3: Design concepts explained

Video: https://www.youtube.com/watch?v=BWhh3r6W-qE
Blog: https://explosion.ai/blog/spacy-design-concepts

spaCy is a popular open-source library for industrial-strength Natural Language Processing in Python. spaCy v3.0 features new transformer-based pipelines that get spaCy’s accuracy right up to the current state-of-the-art, and a new training config and workflow system to help you take projects from prototype to production. In this video, I’ll show you some of the new design concepts and explain what’s going on under the hood, how we’ve implemented them and most importantly, why. I’ll also share some lessons we’ve learned about developer experience along the way.

Transcript

1. v3 @_inesmontani NEW LIBRARY PATTERNS & Behind the Scenes: Design Concepts EXPLAINED
3. global function registry system
4. global function registry system programmable user-facing APIs
5. global function registry system programmable user-facing APIs “bottom-up” configuration system
6. global function registry system programmable user-facing APIs type-based data validation “bottom-up” configuration system
7. global function registry system programmable user-facing APIs type-based data validation “bottom-up” configuration system type hints & static analysis for model definitions
9. advanced workflows for modern NLP & deep learning SCENARIO #1
10. advanced workflows for modern NLP & deep learning SCENARIO #1 ease of use with pre-configured building blocks & good defaults SCENARIO #2
11. Machine Learning is complex AND THAT’S OKAY
12. Developer Experience NOT JUST WE NEED BETTER Abstractions
13. custom registered functions and code config overrides (data paths) spacy.io/usage/training training config
14. con fi g.cfg customize settings, hyperparameters and architectures
15. docs.python.org/3/library/con fi gparser.html
16. docs.python.org/3/library/con fi gparser.html + any JSON values + more flexible variable interpolation
17. con fi g.cfg registered functions resolved bottom-up variable interpolation structured sections
18. “classic” approach
19. modular approach
20. con fi g.cfg
21. con fi g.cfg resolved bottom-up
22. con fi g.cfg resolved bottom-up
23. Pseudocode “top-down” configuration
24. Pseudocode “top-down” configuration
25. Pseudocode “top-down” configuration
26. Pseudocode “top-down” configuration
27. Pseudocode “top-down” configuration
28. Pseudocode “bottom-up” configuration
29. Pseudocode “bottom-up” configuration
30. con fi g.cfg
31. con fi g.cfg
32. serialized model save
33. serialized model save load
34. serialized model save load custom code & settings
35. serialized model save load custom code & settings How should I reconstruct this object?
36. serialized model save load custom code & settings How should I reconstruct this object? define how to create custom objects
37. github.com/explosion/catalogue Function registry
38. github.com/explosion/catalogue Function registry
39. github.com/explosion/catalogue Function registry
40. we always need to know how an object expects to be created spacy.io/usage/processing-pipelines
41. docs.python.org/3/library/typing.html Python 3 syntax feature
42. docs.python.org/3/library/typing.html Python 3 syntax feature expected types
43. Bugs & Mistakes HAPPEN – WE JUST NEED TO catch them
44. github.com/samuelcolvin/pydantic data validation and settings management using Python type hints
46. define data model
47. define data model validate against data model
48. define data model validate against data model ... FilePath HttpUrl int bool StrictStr PositiveInt
50. catch errors in config
51. Con fi g schema
52. Con fi g schema con fi g.cfg
53. Con fi g schema con fi g.cfg
55. 1. inspect
56. 1. inspect 2. generate data model
57. 1. inspect 2. generate data model 3. validate
58. base_con fi g.cfg
59. base_con fi g.cfg partial config
60. base_con fi g.cfg partial config show visual diff con fi g.cfg
75. Debugging & FASTER Productivity HIGHER
76. Model thinc.ai
77. Model custom array types Floats2d thinc.ai
78. Model custom array types Floats2d Ints1d ... Padded Ragged thinc.ai
80. expected return types
81. Y: Floats3d Incompatible return value type (got "Tuple[Floats3d, Callable[[Any], Any]]", expected return types
82. Y: Floats3d Incompatible return value type (got "Tuple[Floats3d, Callable[[Any], Any]]", expected return types static analysis: catch errors as you type
83. Pseudocode typed methods for transformations
84. Model thinc.ai
85. Model Model[InputT, OutputT] generic types thinc.ai
87. mypy.ini optional mypy plugin for more checks
88. Relu: Relu Layer outputs type (thinc.types.Floats2d) but the next layer expects (thinc.types.Ragged) as an input mypy.ini optional mypy plugin for more checks
89. Relu: Relu Layer outputs type (thinc.types.Floats2d) but the next layer expects (thinc.types.Ragged) as an input static analysis: catch errors as you type mypy.ini optional mypy plugin for more checks
90. Developer Productivity
91. consistent naming Developer Productivity
92. user-focused error handling & validation consistent naming Developer Productivity
93. extensive documentation user-focused error handling & validation consistent naming Developer Productivity
94. extensive documentation user-focused error handling & validation consistent naming avoid redundant shortcuts & competing abstractions Developer Productivity
95. extensive documentation user-focused error handling & validation consistent naming avoid redundant shortcuts & competing abstractions smooth path from prototype to production Developer Productivity
96. extensive documentation user-focused error handling & validation consistent naming avoid redundant shortcuts & competing abstractions smooth path from prototype to production provide building blocks to program with, not just abstractions Developer Productivity
97. extensive documentation user-focused error handling & validation consistent naming avoid redundant shortcuts & competing abstractions smooth path from prototype to production provide building blocks to program with, not just abstractions Developer Productivity
98. Prototype & Production CLOSING THE GAP BETWEEN
99. NLP
100. NLP flexible tools you know and understand
101. spacy.io/usage/v3 @spacy_io @_inesmontani
102. spacy.io/usage/v3 @spacy_io @_inesmontani install spaCy v3 from pip or conda
103. spacy.io/usage/v3 @spacy_io @_inesmontani documentation and quickstart install spaCy v3 from pip or conda
104. spacy.io/usage/v3 @spacy_io @_inesmontani documentation and quickstart install spaCy v3 from pip or conda thank you! —