Machine Learning Seminar (1)

Supervised Learning Basics

Transcript

1.  (1)
     April 17, 2019 Tomoya Nakayama
2.  
    • "/!-%+!- ' • ,. • 2(*  , • #(* •   • &) 
    →$  2
3. 1. 
    3
4.   x y 1 3 2 5 3 7 4 9 5 11 6 ? • ! = #$ + &  wb
   • #
   (weight) • &(bias) 4
5. 
   (1) 1. w b+  • w
   b0 $ 2. w b&
   (x  y #,*"' • ! "# = %&# + ( (0 ≤ , ≤ -)
   n )! 0 2 4 6 8 10 12 0 1 2 3 4 5 6 y x % * #,* 5 w=1, b=0 
6.  (2) 3. !   2
    x     (Mean Squared Error) • ! = # $ ∑ &'# $ () *& − *& )- ↑ 
   (Loss Function) 0 2 4 6 8 10 12 0 1 2 3 4 5 6 y x   !  2  (7-3)^2=16 6
7. 
   (3) 4. &%!xb  +* " )J’(x), J’(b)  • !" # = % %& ! • !" ' = % %( ! 5. J’(x), J’(b)(learning rate)α$  x, b     x, b  • # ← # − +!" # • ' ← ' − +!′(') 6. 25'(#
   7
8. 
    x1 x2 y 1 0 3 2 1 6 3 0 7 4 1 10 5 0 11 6 1 ? • ! = #$ %$ + #' %' + (  w1, w2, b • ) = %$ %' , + = #$ #'    
   , = -./ + 0  •   !   8
9. 2 1 2
    64 75
   77 66  82 55
   49 95  58 83  69 69 ? 40 50 60 70 80 90 100 40 50 60 70 80 90 9
10. 2 1 2
     64 75
    77 66  82 55
    49 95  58 83  69 69 ? 40 50 60 70 80 90 100 40 50 60 70 80 90 10     
11. 2  
     • 27!#1;#0'71 :  • +6"-1$"-0 • 1.3)0 • 4 .3&,710=895 • .3 0.5(  1 0 *< • 0.5>7/%2  11
12. 2 
     • ! " = $(&'( + *) • σ (sigmoid) • 01  !  • 1 1 
    0 0 ! W b  0 0.5 1 -6 -4 -2 0 2 4 6 y x  12
13. 2  • ! = # $ ∑ &'# $ −)& log - )& − 1 − )& log 1 − - )& • ↑ 
    (cross entropy) • σ 
      13
14. 2  • %+*(accuracy)#  • 
    %'10   .  • "!1/0&- %+*  • ,*($* •  (precision)…1&-   %'1) •  (recall)…%'1 1&- ) 
     1 (Positive) 0 (Negative) 1 (Positive) True Positive False Positive 0 (Negative) True Negative False Negative 14
15. 
     • "3   • ! " = $%&'()*(,-. + 0) • , = 233 234 235 243 ⋱ ⋮ 253 … ⋱ ∈ ℝ;×=, 0 = ?3 ?4 ?5 ∈ ℝ=, ! " ∈ ℝ= • $%&'()* *@ = ABC ∑ CEF G ABG • ( 
    H"  • Softmax!  $" • 2!" #  15
16. 
    (MLP) • Y=WTX+b),4&.4&3(
    2-/$'*  • 3  :hidden layer)ReLU +%# &" 16 ←1033MLP !33 3534 !32
17. 
     • .4 "EC"!#"- • 8>69(cross validation)1"7  • EC"!#"3@:
    ?=; • %' !,2A  • 3B0(regularization)…($+!D</ • ,' !,2A   • "5 • &*)5 17
18. 2. 
     18
19.   
    
     19
20. 
     • numpy
     • scikit-learn
       • pandas
      • matplotlib
      • seaborn
      20
21. 
     • Jupyter Notebook/Lab •
       • JupyterIDE • PyCharm • Visual Studio Code • Anaconda • PCpython   21
22. 
     • Amazon SageMaker • Azure Notebook • Azure
      • Google Colaboratory • Google
     • Binder • GitHub/GitLabJupyter Notebook  22
23. 3. 
     23
24. 
     1. *,!/0 • Google Colab4 • Binder4 2. MNIST)2&.0 • Google Colab4 • Binder4 • Google($  Google Colab4#3   • Google($  Binder4#3   • Binder4+%10-" 
     • Binder4GPU(  %'1 24