我如何在PyCon找到乾爹 ─ 工人智慧的展現

Lightning Talk @ COSCUP 2018

Transcript

1. ౯ই֜ࣁPyConತکԦᇂ ʒ ૡՈฬ眻ጱ疻匍 Bingroom
2. 獋㮆์獮
3. 秚瑊䋊聜೴䜗
4. 秚瑊䋊聜೴䜗
5. 秚瑊䋊聜೴䜗 Task: 狕硬蟈կ獉਻牧ֵ秇ࣳ藮ڣ牧蝢螂࣯璀蟈կ䲒介
6. 秚瑊䋊聜೴䜗 Task: 狕硬蟈կ獉਻牧ֵ秇ࣳ藮ڣ牧蝢螂࣯璀蟈կ䲒介
7. Run sample code
10. Run sample code (cont’d) • ౯䷱磪python3
 => 箑ԧ碉㮆pycon day1 • ౯䷱磪pip3
 => python3 -mpip [module]
11. Run sample code (cont’d) • ౯䷱磪python3
 => 箑ԧ碉㮆pycon day1 • ౯䷱磪pip3
 => python3 -mpip [module]
12. Run sample code (cont’d)
13. Run sample code (cont’d)
14. result = 1: ᤩ瞟کԧ牧ฎ࣯璀蟈կ牦 result = 0: 覍࣯璀蟈կ牧௲ࡅ spamڣ獨ሲ result
15. • 褰׎揳Ӟ缰෈ᒍ荠ധ࣯璀蟈կ獉਻
16. Mission Complete • 褰׎揳Ӟ缰෈ᒍ荠ധ࣯璀蟈կ獉਻
17. Mission Complete? Task 2: 吚磪ग़֖㷢搴ᘏ藮ڣሲፘݶ碻牧㳷穉斃翥蜉蚣櫝 (Edit Distance)牧 犥翥蜉蚣櫝斃ੜጱ㷢搴ᘏ㵛ڊ牐 • 褰׎揳Ӟ缰෈ᒍ荠ധ࣯璀蟈կ獉਻
18. Edit Distance
19. Keras
20. Keras • 碝ಋ㴄ሻӞ㮆์ • print(model.summary())
 => ฎ඘裾ԫز獤觊ጱ๏݋珌
21. Keras • 碝ಋ㴄ሻӞ㮆์ • print(model.summary())
 => ฎ඘裾ԫز獤觊ጱ๏݋珌
22. • print (tokenizer.word_index) Idea 1 - 猂፡ਁَ vocab_size:147508
23. • shuffle letters of word 玲դܻਁ ex: hello -> hleol • 虋୵: stemming / +s / +ing ex: code -> coding • ೪ṛӾ舣෈㻌ਁᰁ ex: Subject -> Suspect 
 (Edit distance = 2) Idea 2 - ᛔୌmapping
24. • shuffle letters of word 玲դܻਁ ex: hello -> hleol • 虋୵: stemming / +s / +ing ex: code -> coding • ೪ṛӾ舣෈㻌ਁᰁ ex: Subject -> Suspect 
 (Edit distance = 2) Idea 2 - ᛔୌmapping
25. Framework
26. Framework • 櫝 submit ݝ獵 2hr
 • 獊ಋ㵕䋿匍: Ոૡฬ眻 -> ૡՈฬ眻

27. Framework • 櫝 submit ݝ獵 2hr
 • 獊ಋ㵕䋿匍: Ոૡฬ眻 -> ૡՈฬ眻

28. Manual Strategy • spam word -> frequent word
 => i -> i , the -> they • change / delete single letter
 => only -> ondy , control -> contro , spam -> adam
 hey -> hy , tried -> fried , wish -> ish • य़ᰁ +s / +ed / +ing
 => program -> programs , windows -> windowss , 
 address -> addressed
29. Manual Strategy • spam word -> frequent word
 => i -> i , the -> they • change / delete single letter
 => only -> ondy , control -> contro , spam -> adam
 hey -> hy , tried -> fried , wish -> ish • य़ᰁ +s / +ed / +ing
 => program -> programs , windows -> windowss , 
 address -> addressed
30. Manual Strategy • spam word -> frequent word
 => i -> i , the -> they • change / delete single letter
 => only -> ondy , control -> contro , spam -> adam
 hey -> hy , tried -> fried , wish -> ish • य़ᰁ +s / +ed / +ing
 => program -> programs , windows -> windowss , 
 address -> addressed
31. Manual Strategy (cont’d) • Use statistically rare letter (ex: X, W ,…)
 => xp -> x , nt -> nw , way -> www • Concern on meaningful numbers
 => 2000 -> 200 , 60 -> 6 , 9 -> 9w • ೪ṛӾ舣෈㻌ਁᰁ ex: Subject -> Suspect 
 (Edit distance = 2)
32. Manual Strategy (cont’d) • Use statistically rare letter (ex: X, W ,…)
 => xp -> x , nt -> nw , way -> www • Concern on meaningful numbers
 => 2000 -> 200 , 60 -> 6 , 9 -> 9w • ೪ṛӾ舣෈㻌ਁᰁ ex: Subject -> Suspect 
 (Edit distance = 2)
33. Manual Strategy (cont’d) • Use statistically rare letter (ex: X, W ,…)
 => xp -> x , nt -> nw , way -> www • Concern on meaningful numbers
 => 2000 -> 200 , 60 -> 6 , 9 -> 9w • ೪ṛӾ舣෈㻌ਁᰁ ex: Subject -> Suspect 
 (Edit distance = 2)
34. Evaluation
35. Result
36. Result 碻樌姴蝁牧᩼㬵᩼窾 (य़腭)
37. Discussion • How can I win?
38. Discussion • How can I win? 穉搴㷢ےՈ碍 < 10
39. ತکԦᇂԧ牦 Ԧᇂ Ԧᇂ Ԧᇂ
40. THE END