Протодьяконов Алгоритмы Data Science и их практическая реализация на Python 2022
.pdf
ǶǻȀdzǾǽǼǹȍȄǶȍ Ƕ ȋǸǿȀǾǮǽǼǹȍȄǶȍ
ɉɨɫɬɚɧɨɜɤɚ ɢɫɯɨɞɧɨɣ ɡɚɞɚɱɢ ɉɨɫɬɪɨɢɬɶ ɦɨɞɟɥɶ ɷɧɟɪɝɨɩɨɬɪɟɛɥɟɧɢɹ ɡɞɚɧɢɹ ɩɨ ɱɚɫɚɦ ɉɨɝɨɞɭ ɢ ɯɚɪɚɤ
ɬɟɪɢɫɬɢɤɢ ɡɞɚɧɢɹ ɩɨɤɚ ɧɟ ɪɚɫɫɦɚɬɪɢɜɚɬɶ Ⱦɚɧɧɵɟ KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH WUDLQ FVY J]
Ɂɚɝɪɭɡɤɚ ɛɢɛɥɢɨɬɟɤ ȼɤɚɱɟɫɬɜɟɩɟɪɜɨɝɨɲɚɝɚɥɨɝɢɱɧɨɢɰɟɥɟɫɨɨɛɪɚɡɧɨɡɚɝɪɭɡɢɬɶɛɢɛɥɢɨɬɟɤɢ
Ɂɚɝɪɭɡɤɚ ɞɚɧɧɵɯ Ɂɚɬɟɦ ɩɟɪɟɯɨɞɢɦ ɤ ɡɚɝɪɭɡɤɟ ɧɟɩɨɫɪɟɞɫɬɜɟɧɧɨ ɢɫɯɨɞɧɵɯ ɞɚɧɧɵɯ
Ɉɛɨɝɚɳɟɧɢɟ ɞɚɧɧɵɯ Ⱦɨɛɚɜɢɦ ɫɟɪɢɸ ɫ ɱɚɫɨɦ ɫɭɬɨɤ ɞɥɹ ɩɨɫɬɪɨɟɧɢɹ ɫɭɬɨɱɧɨɣ ɦɨɞɟɥɢ ɩɨɬɪɟɛ
ɥɟɧɢɹ
50
ɋɪɟɞɧɟɟ ɩɨɬɪɟɛɥɟɧɢɟ ɩɨ ɱɚɫɚɦ ȼɵɜɟɞɟɦ ɫɪɟɞɧɟɟ ɢ ɦɟɞɢɚɧɭ ɩɨɬɪɟɛɥɟɧɢɹ ɷɧɟɪɝɢɢ ɩɨ ɱɚɫɚɦ
ɇɚ ɜɵɯɨɞɟ ɪɚɛɨɬɵ ɹɱɟɣɤɢ ɛɭɞɟɬ ɩɨɥɭɱɟɧ ɫɥɟɞɭɸɳɢɣ ɝɪɚɮɢɤ
Ɏɢɥɶɬɪɭɟɦ ɦɟɬɪɢɤɭ ɍɞɚɥɹɟɦ ɧɭɥɟɜɵɟ ɡɧɚɱɟɧɢɹ ɢɡ ɫɬɚɬɢɫɬɢɤɢ
ɉɨɥɭɱɚɹ ɫɥɟɞɭɸɳɢɣ ɪɟɡɭɥɶɬɚɬ ɮɭɧɤɰɢɨɧɢɪɨɜɚɧɢɹ:
51
ɂɧɬɟɪɩɨɥɢɪɭɟɦ ɞɚɧɧɵɟ ɩɨ ɱɚɫɚɦ ɉɨɫɬɪɨɢɦ ɦɨɞɟɥɶ ɜɧɭɬɪɢɫɭɬɨɱɧɨɝɨ ɩɨɬɪɟɛɥɟɧɢɟ ɷɧɟɪɝɢɢ ɩɨ ɡɞɚɧɢɸ
Ɍɟɩɟɪɶ ɝɪɚɮɢɤ ɛɭɞɟɬ ɢɦɟɬɶ ɬɚɤɨɣ ɜɢɞ
ǼȄdzǻǸǮ ǺǼDzdzǹǶ
ɉɨɫɬɚɧɨɜɤɚ ɡɚɞɚɱɢ ɉɨɫɬɪɨɢɬɶ ɩɪɨɫɬɭɸ ɦɨɞɟɥɶ ɷɧɟɪɝɨɩɨɬɪɟɛɥɟɧɢɹ ɡɞɚɧɢɹ ɩɨ ɫɪɟɞɧɟɦɭ ɡɧɚ
ɱɟɧɢɸ ɨɰɟɧɢɬɶ ɷɮɮɟɤɬɢɜɧɨɫɬɶ ɦɨɞɟɥɢ ɱɟɪɟɡ ɦɟɬɪɢɤɭ
|
n |
RMSLE |
¦ ORJ pi ORJ ai 2 |
|
i 1
n
ɝɞɟ n – ɱɢɫɥɨ ɧɚɛɥɸɞɟɧɢɣ ORJ – ɧɚɬɭɪɚɥɶɧɵɣ ɥɨɝɚɪɢɮɦ
pi – ɜɵɱɢɫɥɟɧɧɨɟ ɡɧɚɱɟɧɢɟ ɦɟɬɪɢɤɢ ai – ɡɚɞɚɧɧɨɟ ɡɧɚɱɟɧɢɟ ɦɟɬɪɢɤɢ
Ⱦɚɧɧɵɟ KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH WUDLQ FVY J]
Ɂɚɝɪɭɡɤɚ ɛɢɛɥɢɨɬɟɤ Ⱦɨɩɨɥɧɢɬɟɥɶɧɨ ɫɪɚɡɭ ɨɬɫɟɱɟɦ ɩɭɫɬɵɟ ɞɧɢ ɢ ɜɵɞɟɥɢɦ ɱɚɫ ɢɡ ɡɧɚɱɟɧɢɹ
ɜɪɟɦɟɧɢ
52
Ɂɚɝɪɭɡɤɚ ɞɚɧɧɵɯ Ⱦɨɩɨɥɧɢɬɟɥɶɧɨ ɫɪɚɡɭ ɨɬɫɟɱɟɦ ɩɭɫɬɵɟ ɞɧɢ ɢ ɜɵɞɟɥɢɦ ɱɚɫ ɢɡ ɡɧɚɱɟɧɢɹ
ɜɪɟɦɟɧɢ
Ɋɚɡɞɟɥɟɧɢɟ ɞɚɧɧɵɯ ɧɚ ɨɛɭɱɟɧɢɟ ɢ ɩɪɨɜɟɪɤɭ ȼɵɞɟɥɢɦ % ɜɫɟɯ ɞɚɧɧɵɯ ɧɚ ɩɪɨɜɟɪɤɭ ɨɫɬɚɥɶɧɵɟ ɨɫɬɚɜɢɦ ɧɚ ɨɛɭɱɟ
ɧɢɟ
ɋɨɡɞɚɞɢɦ ɦɨɞɟɥɢ ɋɪɟɞɧɟɟ ɢ ɦɟɞɢɚɧɧɨɟ ɡɧɚɱɟɧɢɟ ɩɨɬɪɟɛɥɟɧɢɟ ɷɧɟɪɝɢɢ ɩɨ ɱɚɫɚɦ
53
Ɏɭɧɤɰɢɹ ɩɪɨɜɟɪɤɢ ɦɨɞɟɥɢ
|
n |
|
RMSLE |
¦ ORJ pi ORJ ai 2 |
. |
i 1 |
||
|
n |
|
Ⱦɥɹ ɜɵɱɢɫɥɟɧɢɹ ɦɟɬɪɢɤɢ ɫɨɡɞɚɞɢɦ ɲɟɫɬɶ ɧɨɜɵɯ ɫɬɨɥɛɰɨɜ ɜ ɬɟɫɬɨɜɨɦ ɧɚɛɨɪɟ ɞɚɧɧɵɯ ɫ ɥɨɝɚɪɢɮɦɨɦ ɡɧɚɱɟɧɢɹ ɦɟɬɪɢɤɢ ɩɪɟɞɫɤɚɡɚɧɢɟɦ ɩɨ ɫɪɟɞɧɟɦɭ ɢ ɩɨ ɦɟɞɢɚɧɟ ɚ ɬɚɤɠɟ ɫ ɤɜɚɞɪɚɬɨɦ ɪɚɡɧɢɰɵ ɩɪɟɞɫɤɚɡɚɧɢɣ ɢ ɥɨɝɚɪɢɮɦɚ ɡɧɚɱɟ ɧɢɹ ɉɨɫɥɟɞɧɢɣ ɫɬɨɥɛɟɰ ɞɨɛɚɜɢɦ ɱɬɨɛɵ ɫɪɚɜɧɢɬɶ ɩɪɟɞɫɤɚɡɚɧɢɟ ɫ ɟɝɨ ɨɬɫɭɬ ɫɬɜɢɟɦ – ɧɭɥɹɦɢ ɜ ɡɧɚɱɟɧɢɹɯ
54
Ɍɟɩɟɪɶ ɨɫɬɚɟɬɫɹ ɩɪɨɫɭɦɦɢɪɨɜɚɬɶ ɤɜɚɞɪɚɬɵ ɪɚɫɯɨɠɞɟɧɢɣ ɪɚɡɞɟɥɢɬɶ ɧɚ ɤɨɥɢɱɟɫɬɜɨ ɡɧɚɱɟɧɢɣ ɢ ɢɡɜɥɟɱɶ ɤɜɚɞɪɚɬɧɵɣ ɤɨɪɟɧɶ
ǹǶǻdzǷǻǮȍ ǾdzDZǾdzǿǿǶȍ
ɉɨɫɬɚɧɨɜɤɚ ɡɚɞɚɱɢ ɉɨɫɬɪɨɢɬɶ ɦɨɞɟɥɶ ɥɢɧɟɣɧɨɣ ɪɟɝɪɟɫɫɢɢ ɷɧɟɪɝɨɩɨɬɪɟɛɥɟɧɢɹ ɡɞɚɧɢɹ ɢɫ
ɩɨɥɶɡɭɹ ɬɟɦɩɟɪɚɬɭɪɭ ɜɨɡɞɭɯɚ DLUBWHPSHUDWXUH ɢ ɜɥɚɠɧɨɫɬɶ GHZBWHPSHUDWXUH Ɋɚɫɫɱɢɬɚɬɶ ɤɚɱɟɫɬɜɨ ɩɨɫɬɪɨɟɧɧɨɣ ɦɨɞɟɥɢ ɩɨ ɩɪɨɜɟɪɨɱɧɵɦ ɞɚɧɧɵɦ Ⱦɚɧɧɵɟ
1.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH EXLOGLQJBPHWDGDWD FVY J]
2.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH ZHDWKHUBWUDLQ FVY J]
3.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH WUDLQ FVY J]
ɉɨɞɤɥɸɱɟɧɢɟ ɛɢɛɥɢɨɬɟɤ
Ɂɚɝɪɭɡɤɚ ɞɚɧɧɵɯ
55
Ɉɛɴɟɞɢɧɟɧɢɟ ɞɚɧɧɵɯ ɢ ɮɢɥɶɬɪɚɰɢɹ
Ⱦɨɛɚɜɥɟɧɢɟ ɱɚɫɚ ɜ ɞɚɧɧɵɟ
Ɋɚɡɞɟɥɟɧɢɟ ɞɚɧɧɵɯ ɧɚ ɨɛɭɱɟɧɢɟ ɢ ɩɪɨɜɟɪɤɭ
56
Ɇɨɞɟɥɶ ɥɢɧɟɣɧɨɣ ɪɟɝɪɟɫɫɢɢ ɢ ɫɪɟɞɧɟɟ
PHWHUBUHDGLQJ $ DLUBWHPSHUDWXUH % GHZBWHPSHUDWXUH & Ⱦɨɩɨɥɧɢɬɟɥɶɧɨ ɜɵɱɢɫɥɢɦ ɫɪɟɞɧɟɟ ɩɨ ɱɚɫɚɦ ɱɬɨɛɵ ɫɪɚɜɧɢɬɶ ɥɢɧɟɣɧɭɸ
ɪɟɝɪɟɫɫɢɸ ɫ ɛɨɥɟɟ ɩɪɨɫɬɨɣ ɦɨɞɟɥɶɸ
Ɉɰɟɧɤɚ ɦɨɞɟɥɢ
ǼǽȀǶǺǶǵǮȄǶȍ ǽǼȀǾdzǯǹdzǻǶȍ ǽǮǺȍȀǶ
ɉɨɫɬɚɧɨɜɤɚ ɡɚɞɚɱɢ Ɂɚɝɪɭɡɢɬɶ ɞɚɧɧɵɟɩɨ ɷɧɟɪɝɨɩɨɬɪɟɛɥɟɧɢɸ ɜɫɟɯ ɡɞɚɧɢɣ ɜ ɨɩɟɪɚɬɢɜɧɭɸɩɚ
ɦɹɬɶ ɢ ɞɨɛɢɬɶɫɹ ɟɟ ɦɢɧɢɦɚɥɶɧɨɝɨ ɪɚɫɯɨɞɚ Ⱦɚɧɧɵɟ
1.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH EXLOGLQJBPHWDGDWD FVY J]
2.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH ZHDWKHUBWUDLQ FVY J]
3.KWWS YLGHR LWWHQVLYH FRP PDFKLQH OHDUQLQJ DVKUDH WUDLQ FVY J]
57
ɉɨɞɤɥɸɱɟɧɢɟ ɛɢɛɥɢɨɬɟɤ
Ɍɨɱɧɨɫɬɶ ɢ ɪɚɡɦɟɪ ɬɢɩɨɜ
Ɂɚɝɪɭɡɤɚ ɞɚɧɧɵɯ
58
ɉɨɬɪɟɛɥɟɧɢɟ ɩɚɦɹɬɢ
Ɏɭɧɤɰɢɹ ɨɩɬɢɦɢɡɚɰɢɹ ɩɚɦɹɬɢ
Ɉɩɬɢɦɢɡɚɰɢɹ ɩɚɦɹɬɢ ɫɬɪɨɟɧɢɹ
59