Formula 1 and the same plate with the speed of the pilots

Occasion

A couple of months ago, as part of the collaboration between Amazon and Formula 1, researchers "using the power of cloud technologies" rolled out a comparison of the speed of pilots of all time ( link ). Naturally, the material was hype advertising, and it achieved its goal. In the whole formula world for a couple of days, there was only talk in the spirit of "why is there no pilot N in the rating?" and "how can M be faster than K if K made it in season L." It became interesting for me to more or less repeat this research and, if possible, do without "all the power of cloud technologies."

A task

. . , , , , , . . AWS – . , . .

F1, – 10% , 90% - . - , .. , , . , , . , , .

. , , «» . , 2 , , « » - , . , . , , , .

. , , . «» , , 1. , , , .

, , k_p, .

.. V = V₀*k_p*k_c*Tr, k_c – , Tr – , .

, ln(k_i)-ln(k_j)=ln(t_j)-ln(t_i) .

. m*(n+1), n – , m – . 1 -1 , , .

AWS, , , - . , , « » - , . .. «» 2 0 .

, F1. . , , , , . , , , - .

  , , 2 . 3 . 1 – , . , «». – . , . . – . , , .

, . , 1, 2 3 . , , .

import sys
import re
import urllib.request


def get_wikipedia_page(title, lang='en'):
    url = 'https://'+lang+'.wikipedia.org/wiki/'+(title.replace(' ', '_'))
    fp = urllib.request.urlopen(url)
    mybytes = fp.read()
    
    mystr = mybytes.decode("utf8")
    fp.close()
    return mystr

title = 'List of Formula One Grands Prix'

try:
    print('process: '+title)
    th = get_wikipedia_page(title)
    r1 = re.findall(r'href="/wiki/[\d][\d][\d][\d]_[\w]*_Grand_Prix"',th)
    list_of_GP = list(set(r1))
except:
    print("Unexpected error:", sys.exc_info()[0])

titles = list(map(lambda x: x[12: -1].replace('_', ' '), list_of_GP))


for title in titles:
    try:
        print('process: '+title)
        th = get_wikipedia_page(title)
        with open('texts/'+title+'.txt', 'w', encoding='utf8') as the_file:
            the_file.write(th)
            the_file.close()
    except:
        print("Unexpected error:", sys.exc_info()[0])

html . . ( ), ( ), «» . csv.

DataFrame. , :

+ . ,

+ – , 1 , 2

= + . , . , . - , .

qual_df = pd.read_csv('qual_results.csv')

qual_df['Track_pl_len'] = qual_df['Track'] + '_' +qual_df['Track_len'].apply(str)
qual_df['Car'] = qual_df['Constructor'] + '_' +qual_df['Year'].apply(str)
qual_df['Driver_pl_year'] = qual_df['Driver']+'_'+qual_df['Year'].apply(str)

qual_df_2 = qual_df.copy()
qual_df_2['Driver_pl_year'] = qual_df_2['Driver']+'_'+((qual_df_2['Year'].apply(int)-1)).apply(str)
double_df = pd.concat([qual_df, qual_df_2])
del qual_df2

. , . . , , . , , . .

. 2 , .

• , ..

• , 2 –

• , «» «» - 6 2%. , «» 1 – , . ,

• 2 One Hot Encoding , x = 1 , y -1

• , . , – 1 , .. 1, 0. –

• . ln(k) - .

, .

, , «» , .. , F1 Amazon.

, :

,

18 , . , + + , . , 10. , . . , .

– . 2019 , . , 1.

– . . , . , 1 . . 1, / .

, :

  ( , .. 1 ):

  , .

:

2020 – , ( – 1979 , ).

, Renault , . Racing Point, 2 , Alpha Tauri 2020, Red Bull 2019 Ferrari 2020, - 2018 .

, 2019 Mercedes, 10 , Red Bull . , , , , .

. , . - , . . , .

, ( , .. 2 , 1):

1 Amazon . : .

, , , / , , .

, . , , – 70 1. , .

It was not possible to build a decent regression, a seemingly problem-free model still throws up unexpected results. All attempts to construct a unified model, taking into account both the influence of the machine and the pilot, have come to the necessity of strict regularization.

I attach a github to the article with initial data and pivot tables for self-study, I'll add the code as soon as I'm not so ashamed of it.

→ Link to Github