VanillaJS ML for Web Developers

Bridge JS development and Genetic Algorithm.
+ Demo: https://evoria-1984.firebaseapp.com
+ GitLab: https://gitlab.com/JBClion/Evoria

Transcript

1. Evoria VanillaJS ML for Web Developers Jean-Baptiste Clion Google Developer Expert
2. #dfua My today objective, Communicate my interest for Genetic Algorithms. This talk would be successful if I can convince you to try or to write a GA. - Let’s Get Started !
3. ACT I FREE BOB!
12. EVORIA
13. SUMMARY & CHECK POINT
14. ACT II Encoding, Genotype, Phenotype and Searchspace
15. 4 possible movements
16. Encoded in 2 bits 22 = 4 possibilities 00 01 10 11
17. 16 possible configurations
18. 0000 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111 Encoded in 4 bits, 24 = 16
19. 4 possible movements for each of the 16 possible configurations
20. 1 solution = 1 Species
21. 1 new species 1 mutation
22. 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111 0000 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01
23. Encoding The process of converting data into a format required for processing, computing or combinatory needs.
24. 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111 0000 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01 Genotype Phenotype
25. 0001 0010 0011 0100 0101 0110 0111 1000 1001 1010 1011 1100 1101 1110 1111 0000 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01
26. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 0 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01
27. 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01
28. 10 00 11 00 10 01 00 01 00 11 01 00 01 01 01 01
30. 01100011001001000100110100010101 32 bits = 232 = 4,294,967,296 possibilities four billion, two hundred and ninety-four million, nine hundred and sixty-seven thousand, two hundred and ninety-six
31. 00000000000000000000000000000000 TO 11111111111111111111111111111111
32. Search space The set of all possible solutions able to satisfy a given problem’s constraints.
33. SUMMARY & CHECK POINT
34. Scale 1/256
35. Scale 1/256 VALID SOLUTIONS
36. Scale 1/256 VALID SOLUTIONS ? ? ?
37. Intelligence Capacity to search the most adapted solution to a given problem in the most efficient way and within the shortest possible time.
38. Scale 1/256 VALID SOLUTIONS
39. Scale 1/256 VALID SOLUTIONS
40. SUMMARY & CHECK POINT
41. ACT III Fitness, Crossover and Mutation
42. 01010010110000010000110010010000 | 0.24 11001001100011100000111100000011 | 0.48 01010010110111010010110010010000 | 0.64 Fitness operator Execute the genotype to evaluate the toxin How far the toxin can go in the bacteria?
43. 00000000000000000000000000000000 11111111111111111111111111111111 Crossover operator Take 2 possible solutions
44. 00000000000000000000000000000000 11111111111111111111111111111111 00110011001100110011001100110011 Crossover operator Merge them, to jump to another solution
45. 00110011001100110011001100110011 Mutation operator Take a possible solution
46. 00110011001100110011001100110011 00110010001101110011001100100011 Mutation operator Invert bits on a given probability
47. SUMMARY & CHECK POINT
48. CODING TIME!
49. ACT IV Genetic Algorithm
50. 01010010110000010000110010010000 | 0.00 11001001100011100000111100000011 | 0.00 01010010110111010010110010010000 | 0.00 01010010110000010010110010000000 | 0.00 01010010110111010010110010011111 | 0.00 0. Select N random solutions (Here 5) This is the 1st Generation
51. 01010010110111010010110010010000 | 0.64 11001001100011100000111100000011 | 0.48 01010010110000010010110010000000 | 0.44 01010010110111010010110010011111 | 0.32 01010010110000010000110010010000 | 0.24 1. Apply fitness & sort | LOOP START
52. 01010010110111010010110010010000 | 0.64 11001001100011100000111100000011 | 0.48 01010010110000010010110010000000 | 0.44 01010010110111010010110010011111 | 0.32 01010010110000010000110010010000 | 0.24 2. Select the best fit
53. 01010010110111010010110010010000 | 0.64 IF 0.64 >= TARGET THEN STOP ELSE CONTINUE 3. best fit >= target
54. 4. For each genotype in Generation Expect best
55. 01010010110111010010110010010000 4. For each genotype in Generation Expect best
56. 01010010110111010010110010010000 11001001100011100000111100000011 4. For each genotype in Generation Expect best
57. 01010010110111010010110010010000 11001001100011100000111100000011 01000001110011100000111110000011 4. For each genotype in Generation Expect best CROSSOVER WITH BEST
58. 01010010110111010010110010010000 11001001100011100000111100000011 01000001110011100000111110000011 00000001110111100001111110000011 4. For each genotype in Generation Expect best CROSSOVER WITH BEST MUTATE
59. 01010010110111010010110010010000 11001001100011100000111100000011 01000001110011100000111110000011 00000001110111100001111110000011 4. For each genotype in Generation Expect best CROSSOVER WITH BEST MUTATE PUSH TO NEXTGEN
60. 11001001100011100000111100000011 | 0.00 01010010110000010010110010000000 | 0.00 01010010110111010010110010011111 | 0.00 01010010110000010000110010010000 | 0.00 01010010110111010010110010010000 | 0.64 Jump to step 1 4. Add best(s) to new solutions
61. 0. Generate a first Generation 1. Evaluate fitness & sort 2. Select and extract best genotype(s) 3. Is best(s) a matching solutions? Y:6, N:4 4. Crossover each with best(s) & mutate 5. Add best(s) to the new generation, jump to 1 6. Stop Summary
62. SUMMARY & CHECK POINT
63. CODING TIME!
64. ACT V Time to save Bob!
66. ACT VI Understand
67. CROSSOVER
68. Scale 1/256 VALID SOLUTIONS 0.32 0.24 0.44 0.48 0.64 #1 G
69. Scale 1/256 VALID SOLUTIONS 0.32 0.24 0.44 0.48 0.64 #1 G
70. Scale 1/256 VALID SOLUTIONS 0.40 0.60 0.51 0.79 #2 G 0.64
71. Scale 1/256 VALID SOLUTIONS 0.40 0.60 0.51 0.79 #2 G 0.32 0.64 0.24 0.44 0.48
72. Scale 1/256 VALID SOLUTIONS 0.40 0.60 0.51 0.79 #2 G 0.32 0.64 0.24 0.44 0.48 DRIVE THE CONVERGENCE
73. Scale 1/256 VALID SOLUTIONS 0.40 0.60 0.51 0.79 #2 G 0.32 0.64 0.24 0.44 0.48 AVOID REGRESSIONS DRIVE THE CONVERGENCE
74. MUTATION
75. Scale 1/256 VALID SOLUTIONS 0.32 0.24 0.36 0.35 0.18 #1 G
76. Scale 1/256 VALID SOLUTIONS 0.34 0.26 0.36 0.35 0.16 #2 G
77. Scale 1/256 VALID SOLUTIONS 0.32 0.24 0.36 0.37 0.19 #3 G
78. Scale 1/256 VALID SOLUTIONS 0.35 0.29 0.34 0.37 0.21 #4 G
79. Scale 1/256 VALID SOLUTIONS 0.35 0.29 0.34 0.37 0.21 #4 G SIMILAR GENOTYPES
80. Scale 1/256 VALID SOLUTIONS 0.35 0.29 0.34 0.37 0.21 #4 G SIMILAR GENOTYPES = SIMILAR GENERATIONS
81. Scale 1/256 VALID SOLUTIONS 0.35 0.29 0.34 0.37 0.21 #4 G SIMILAR GENOTYPES = SIMILAR GENERATIONS = LOCAL MAXIMA
82. Scale 1/256 VALID SOLUTIONS 0.35 0.29 0.34 0.37 0.21 #4 G SIMILAR GENOTYPES = SIMILAR GENERATIONS = LOCAL MAXIMA = NO CONVERGENCE
83. Scale 1/256 VALID SOLUTIONS 0.35 0.34 0.34 0.37 0.64 #4 G MUTATION ENABLE JUMP ACROSS THE SPACE
84. Scale 1/256 VALID SOLUTIONS 0.35 0.34 0.34 0.37 0.64 #4 G MUTATION ENABLE JUMP ACROSS THE SPACE !! HIGH MUTATION RATIO !! CREATE A RANDOM SEARCH
85. Thank you! Questions? Jean-Baptiste Clion Google Developer Expert Twitter @JBaptisteClion