This page was generated from docs/user_guide/5-marathon_analysis.ipynb. Interactive online version: Binder badge.

Accessing historical data from running race results#

One of the great features in Runpandas is the capability of accessing race’s result datasets accross several races around the world, from majors to local ones (if it’s available at our data repository). In this example, we will explore the results of a famous Major marathon, the 2022 Berlin Marathon and analyze the performance of the WR marathon record from Eliud Kipchoge.

Exploring the results and highlights from the 2022 Berlin Marathon#

More than 30,000 people took the starting line for the 2022 Berlin Marathon, on 25 September 2022. An Elite Platinum Label Marathon, and one of the World marathon Majors. The race is quite famous for its fast and flat course, making it perfect for a record-setting day.

But this particular race also came with special flavours, with the participation of the top runners Eliud Kipchoge in the men’s pro race and American record holder Keira D’Amato in the women’s race.

In this notebook we will explore the results and highlights from the 2022 Berlin Marathon using runpandas methods specially tailored for handling race results data.

Race Overview#

First, let’s load the Berlin Marathon data by using the runpandas method runpandas.get_events. This function provides a way of accessing the race data and visualize the results from several marathons available at our datasets repository. Given the year and the marathon identifier you can filter any marathon datasets that you want analyze. The result will be a list of runpandas.EventData instances with race result and its metadata. Let’s look for Berlin Marathon results.

[194]:

import pandas as pd
import runpandas as rpd
import warnings
warnings.filterwarnings('ignore')

[195]:

results = rpd.get_events('Berlin')
results

[195]:

[<Event: name=Berlin Marathon Results from 2022., country=DE, edition=2022>]

The result comes with the Berlin Marathon Result from 2022. Let’s take a look inside the race event, which comes with a handful method to describe its attributes and a special method to load the race result data into a runpandas.datasets.schema.RaceData instance.

[196]:

berlin_result = results[0]
print('Event type', berlin_result.run_type)
print('Country', berlin_result.country)
print('Year', berlin_result.edition)
print('Name', berlin_result.summary)

Event type RunTypeEnum.MARATHON
Country DE
Year 2022
Name Berlin Marathon Results from 2022.

Now that we confirmed that we requested the corresponding marathon dataset. We will load it into a DataFrame so we can further explore it.

[197]:

#loading the race data into a RaceData Dataframe
race_result = berlin_result.load()
race_result

[197]:
position position_gender country sex division bib firstname lastname club starttime ... 10k 15k 20k 25k 30k 35k 40k grosstime nettime category
0 1 1 KEN M 1 1 Eliud Kipchoge – 09:15:00 ... 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 01:11:08 0 days 01:25:40 0 days 01:40:10 0 days 01:54:53 0 days 02:01:09 0 days 02:01:09 M35
1 2 2 KEN M 1 5 Mark Korir – 09:15:00 ... 0 days 00:28:56 0 days 00:43:35 0 days 00:58:14 0 days 01:13:07 0 days 01:28:06 0 days 01:43:25 0 days 01:59:05 0 days 02:05:58 0 days 02:05:58 M30
2 3 3 ETH M 1 8 Tadu Abate – 09:15:00 ... 0 days 00:29:46 0 days 00:44:40 0 days 00:59:40 0 days 01:14:44 0 days 01:30:01 0 days 01:44:55 0 days 02:00:03 0 days 02:06:28 0 days 02:06:28 MH
3 4 4 ETH M 2 26 Andamlak Belihu – 09:15:00 ... 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 01:11:09 0 days 01:26:11 0 days 01:42:14 0 days 01:59:14 0 days 02:06:40 0 days 02:06:40 MH
4 5 5 KEN M 3 25 Abel Kipchumba – 09:15:00 ... 0 days 00:28:55 0 days 00:43:35 0 days 00:58:14 0 days 01:13:07 0 days 01:28:03 0 days 01:43:08 0 days 01:59:14 0 days 02:06:49 0 days 02:06:49 MH
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
35566 DNF – USA M – 65079 michael perkowski – – ... NaT NaT NaT NaT NaT NaT NaT NaT NaT M65
35567 DNF – USA M – 62027 Karl Mann – – ... NaT NaT NaT NaT NaT NaT NaT NaT NaT M55
35568 DNF – THA F – 27196 oraluck pichaiwongse STATE to BERLIN 2022 – ... NaT NaT NaT NaT NaT NaT NaT NaT NaT W55
35569 DNF – SUI M – 56544 Gerardo GARCIA CALZADA – – ... NaT NaT NaT NaT NaT NaT NaT NaT NaT M50
35570 DNF – AUT M – 63348 Harald Mori Albatros – ... NaT NaT NaT NaT NaT NaT NaT NaT NaT M60

35571 rows × 23 columns

Now you can get some insights about the Berlin Marathon 2022, by using its tailored methods for getting basic and quick insights. For example, the number of finishers, number of participants and the winner info.

[198]:

print('Total participants', race_result.total_participants)
print('Total finishers', race_result.total_finishers)
print('Total Non-Finishers', race_result.total_nonfinishers)

Total participants 35571
Total finishers 34844
Total Non-Finishers 727

[199]:

race_result.winner

[199]:

position                         1
position_gender                  1
country                        KEN
sex                              M
division                         1
bib                              1
firstname                    Eliud
lastname                  Kipchoge
club                             –
starttime                 09:15:00
start_raw_time            09:15:00
half               0 days 00:59:51
5k                 0 days 00:14:14
10k                0 days 00:28:23
15k                0 days 00:42:33
20k                0 days 00:56:45
25k                0 days 01:11:08
30k                0 days 01:25:40
35k                0 days 01:40:10
40k                0 days 01:54:53
grosstime          0 days 02:01:09
nettime            0 days 02:01:09
category                       M35
Name: 0, dtype: object

Eliud Kipchoge of Kenya won the 2022 Berlin Marathon in 2:01:09. Kipchoge’s victory was his fourth in Berlin and 17th overall in a career of 19 marathon starts. And who was the women’s race winner?

[200]:

race_result[(race_result['position_gender'] == 1) & (race_result['sex'] == 'F')].T

[200]:
32
position 33
position_gender 1
country ETH
sex F
division 1
bib F24
firstname Tigist
lastname Assefa
club –
starttime 09:15:00
start_raw_time 09:15:00
half 0 days 01:08:13
5k 0 days 00:16:22
10k 0 days 00:32:36
15k 0 days 00:48:44
20k 0 days 01:04:43
25k 0 days 01:20:48
30k 0 days 01:36:41
35k 0 days 01:52:27
40k 0 days 02:08:42
grosstime 0 days 02:15:37
nettime 0 days 02:15:37
category WH

Tigist Assefa of Ethiopia won the women’s race in a stunning time of 2:15:37 to set a new course record in Berlin.

Runpandas also provides a race’s summary method for showing the compilation of some general insights such as finishers, partipants (by gender and overall).

[201]:

race_result.summary()

[201]:

Event name                    berlin marathon
Event type                                42k
Event country                              DE
Event date                         25-09-2022
Number of participants                  35571
Number of finishers                     34844
Number of non-finishers                   727
Number of male finishers                23314
Number of female finishers              11523
Winner Nettime                0 days 02:01:09
dtype: object

Final Results - The podium finishers#

Now that we have access to all finishers data, we would like to see the podium (top 10) finishers for the men and women’s race. For anwesring this question, we will use Matplotlib, a Python library for creating from basic to advanced plots. Let’s plot the final race time results for men and women.

[202]:

import matplotlib.pyplot as plt

First, let’s pick only the array of the top 10 finishers for men and women.

[203]:

man_top_10 = race_result[(race_result['position_gender'].isin([1,2,3,4,5,6,7,8,9,10])) & (race_result['sex'] == 'M')]
man_top_10

[203]:
position position_gender country sex division bib firstname lastname club starttime ... 10k 15k 20k 25k 30k 35k 40k grosstime nettime category
0 1 1 KEN M 1 1 Eliud Kipchoge – 09:15:00 ... 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 01:11:08 0 days 01:25:40 0 days 01:40:10 0 days 01:54:53 0 days 02:01:09 0 days 02:01:09 M35
1 2 2 KEN M 1 5 Mark Korir – 09:15:00 ... 0 days 00:28:56 0 days 00:43:35 0 days 00:58:14 0 days 01:13:07 0 days 01:28:06 0 days 01:43:25 0 days 01:59:05 0 days 02:05:58 0 days 02:05:58 M30
2 3 3 ETH M 1 8 Tadu Abate – 09:15:00 ... 0 days 00:29:46 0 days 00:44:40 0 days 00:59:40 0 days 01:14:44 0 days 01:30:01 0 days 01:44:55 0 days 02:00:03 0 days 02:06:28 0 days 02:06:28 MH
3 4 4 ETH M 2 26 Andamlak Belihu – 09:15:00 ... 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 01:11:09 0 days 01:26:11 0 days 01:42:14 0 days 01:59:14 0 days 02:06:40 0 days 02:06:40 MH
4 5 5 KEN M 3 25 Abel Kipchumba – 09:15:00 ... 0 days 00:28:55 0 days 00:43:35 0 days 00:58:14 0 days 01:13:07 0 days 01:28:03 0 days 01:43:08 0 days 01:59:14 0 days 02:06:49 0 days 02:06:49 MH
5 6 6 ETH M 2 12 Limenih Getachew – 09:15:00 ... 0 days 00:29:46 0 days 00:44:40 0 days 00:59:42 0 days 01:14:45 0 days 01:30:00 0 days 01:45:06 0 days 02:00:22 0 days 02:07:07 0 days 02:07:07 M30
6 7 7 JPN M 4 15 Kenya Sonota JR East Railway 09:15:00 ... 0 days 00:29:46 0 days 00:44:38 0 days 00:59:39 0 days 01:14:45 0 days 01:30:01 0 days 01:45:06 0 days 02:00:30 0 days 02:07:14 0 days 02:07:14 MH
7 8 8 JPN M 5 24 Tatsuya Maruyama Toyota 09:15:00 ... 0 days 00:30:10 0 days 00:45:28 0 days 01:00:47 0 days 01:15:46 0 days 01:30:43 0 days 01:45:53 0 days 02:01:17 0 days 02:07:50 0 days 02:07:50 MH
8 9 9 JPN M 6 16 Kento Kikutani Toyota Boshoku 09:15:00 ... 0 days 00:29:47 0 days 00:44:39 0 days 00:59:40 0 days 01:14:45 0 days 01:30:02 0 days 01:45:19 0 days 02:01:10 0 days 02:07:56 0 days 02:07:56 MH
9 10 10 KEN M 3 14 Zablon Chumba – 09:15:00 ... 0 days 00:29:45 0 days 00:44:38 0 days 00:59:39 0 days 01:14:45 0 days 01:30:00 0 days 01:45:05 0 days 02:00:49 0 days 02:08:01 0 days 02:08:01 M30

10 rows × 23 columns

[204]:

female_top_10 = race_result[(race_result['position_gender'].isin([1,2,3,4,5,6,7,8,9,10])) & (race_result['sex'] == 'F')]
female_top_10

[204]:
position position_gender country sex division bib firstname lastname club starttime ... 10k 15k 20k 25k 30k 35k 40k grosstime nettime category
32 33 1 ETH F 1 F24 Tigist Assefa – 09:15:00 ... 0 days 00:32:36 0 days 00:48:44 0 days 01:04:43 0 days 01:20:48 0 days 01:36:41 0 days 01:52:27 0 days 02:08:42 0 days 02:15:37 0 days 02:15:37 WH
43 44 2 KEN F 2 F25 Rosemary Wanjiru – 09:15:00 ... 0 days 00:32:38 0 days 00:48:47 0 days 01:04:47 0 days 01:20:54 0 days 01:36:57 0 days 01:53:16 0 days 02:10:10 0 days 02:18:00 0 days 02:18:00 WH
44 45 3 ETH F 3 F8 Tigist Abayechew – 09:15:00 ... 0 days 00:32:38 0 days 00:48:46 0 days 01:04:44 0 days 01:20:49 0 days 01:36:41 0 days 01:52:46 0 days 02:10:15 0 days 02:18:03 0 days 02:18:03 WH
48 49 4 ETH F 1 F5 Workenesh Edesa – 09:15:00 ... 0 days 00:32:37 0 days 00:48:44 0 days 01:04:44 0 days 01:20:48 0 days 01:37:01 0 days 01:53:57 0 days 02:11:15 0 days 02:18:51 0 days 02:18:51 W30
57 58 5 ETH F 4 F6 Sisay Meseret Gola – 09:15:00 ... 0 days 00:32:37 0 days 00:48:44 0 days 01:04:43 0 days 01:20:49 0 days 01:36:42 0 days 01:53:30 0 days 02:12:29 0 days 02:20:58 0 days 02:20:58 WH
60 61 6 USA F 1 F2 Keira D'Amato – 09:15:00 ... 0 days 00:32:43 0 days 00:49:11 0 days 01:05:50 0 days 01:22:36 0 days 01:39:35 0 days 01:57:06 0 days 02:14:28 0 days 02:21:48 0 days 02:21:48 W35
64 65 7 JPN F 5 F18 Rika Kaseda Daihatsu 09:15:00 ... 0 days 00:33:30 0 days 00:50:11 0 days 01:06:56 0 days 01:23:41 0 days 01:40:29 0 days 01:57:27 0 days 02:14:27 0 days 02:21:55 0 days 02:21:55 WH
65 66 8 JPN F 2 F19 Ayuko Suzuki Japan Post Group 09:15:00 ... 0 days 00:33:30 0 days 00:50:10 0 days 01:06:55 0 days 01:23:41 0 days 01:40:30 0 days 01:57:27 0 days 02:14:34 0 days 02:22:02 0 days 02:22:02 W30
66 67 9 JPN F 6 F10 Sayaka Sato Sekisui Chemical 09:15:00 ... 0 days 00:33:17 0 days 00:50:03 0 days 01:06:43 0 days 01:23:29 0 days 01:40:30 0 days 01:57:28 0 days 02:14:48 0 days 02:22:13 0 days 02:22:13 WH
68 69 10 KEN F 7 F7 Vibian Chepkirui – 09:15:00 ... 0 days 00:32:36 0 days 00:48:44 0 days 01:04:43 0 days 01:20:48 0 days 01:36:58 0 days 01:54:12 0 days 02:13:38 0 days 02:22:21 0 days 02:22:21 WH

10 rows × 23 columns

Now, let’s filter only the columns we want to use for the plotting such as the runner’s name and the finish time.

[205]:

male_top_10 =  man_top_10[['firstname', 'lastname', 'nettime']]
male_top_10['fullname'] = male_top_10['firstname'] + ' ' +  male_top_10['lastname']
male_top_10

[205]:
firstname lastname nettime fullname
0 Eliud Kipchoge 0 days 02:01:09 Eliud Kipchoge
1 Mark Korir 0 days 02:05:58 Mark Korir
2 Tadu Abate 0 days 02:06:28 Tadu Abate
3 Andamlak Belihu 0 days 02:06:40 Andamlak Belihu
4 Abel Kipchumba 0 days 02:06:49 Abel Kipchumba
5 Limenih Getachew 0 days 02:07:07 Limenih Getachew
6 Kenya Sonota 0 days 02:07:14 Kenya Sonota
7 Tatsuya Maruyama 0 days 02:07:50 Tatsuya Maruyama
8 Kento Kikutani 0 days 02:07:56 Kento Kikutani
9 Zablon Chumba 0 days 02:08:01 Zablon Chumba 
[206]:

females_top_10 =  female_top_10[['firstname', 'lastname', 'nettime']]
females_top_10['fullname'] = females_top_10['firstname'] + ' ' +  females_top_10['lastname']
females_top_10

[206]:
firstname lastname nettime fullname
32 Tigist Assefa 0 days 02:15:37 Tigist Assefa
43 Rosemary Wanjiru 0 days 02:18:00 Rosemary Wanjiru
44 Tigist Abayechew 0 days 02:18:03 Tigist Abayechew
48 Workenesh Edesa 0 days 02:18:51 Workenesh Edesa
57 Sisay Meseret Gola 0 days 02:20:58 Sisay Meseret Gola
60 Keira D'Amato 0 days 02:21:48 Keira D'Amato
64 Rika Kaseda 0 days 02:21:55 Rika Kaseda
65 Ayuko Suzuki 0 days 02:22:02 Ayuko Suzuki
66 Sayaka Sato 0 days 02:22:13 Sayaka Sato
68 Vibian Chepkirui 0 days 02:22:21 Vibian Chepkirui

Now, we can plot all the data and see the berlin men’s marathon results with visualization of the finishing times.

[207]:

import matplotlib
import datetime


def timeTicks(x, pos):
    seconds = x / 10**9
    d = datetime.timedelta(seconds=seconds)
    return str(d)

fig, ax = plt.subplots()
fig.set_figheight(10)
fig.set_figwidth(15)
ax.barh(male_top_10['fullname'], male_top_10['nettime'],
         edgecolor='grey')
ax.set_yticks(male_top_10['fullname'])
ax.set_yticklabels(male_top_10['fullname'])

formatter = matplotlib.ticker.FuncFormatter(timeTicks)
ax.xaxis.set_major_formatter(formatter)


# show fastest at the top
ax.invert_yaxis()

# draw vertical lines behind the bars
ax.set_axisbelow(True)
ax.xaxis.grid(True, which='major', linestyle='--', color='black', zorder=-1000)


plt.suptitle(f"Berlin Marathon 2022\n" f"Winner finish time: {male_top_10.iloc[0]['nettime']} ({male_top_10.iloc[0]['fullname']})")

plt.show()
../_images/user_guide_5-marathon_analysis_26_0.png 
[208]:

fig, ax = plt.subplots()
fig.set_figheight(10)
fig.set_figwidth(15)
ax.barh(females_top_10['fullname'], females_top_10['nettime'],
         edgecolor='grey')
ax.set_yticks(females_top_10['fullname'])
ax.set_yticklabels(females_top_10['fullname'])

formatter = matplotlib.ticker.FuncFormatter(timeTicks)
ax.xaxis.set_major_formatter(formatter)


# show fastest at the top
ax.invert_yaxis()

# draw vertical lines behind the bars
ax.set_axisbelow(True)
ax.xaxis.grid(True, which='major', linestyle='--', color='black', zorder=-1000)


plt.suptitle(f"Berlin Marathon 2022\n" f"Winner finish time: {females_top_10.iloc[0]['nettime']} ({females_top_10.iloc[0]['fullname']})")

plt.show()
../_images/user_guide_5-marathon_analysis_27_0.png

Runpandas for some race results come with the splits for the partial distances of the race. We can fetch for any runner the splits using the method runpandas.acessors.splits.pick_athlete. So, if we need to have direct access to all splits from a specific runner, we will use the splits acesssor.

[209]:

race_result.splits.pick_athlete(identifier='1')

[209]:
time distance_meters distance_miles
split
0k 0 days 00:00:00 0 0.0000
5k 0 days 00:14:14 5000 3.1069
10k 0 days 00:28:23 10000 6.2137
15k 0 days 00:42:33 15000 9.3206
20k 0 days 00:56:45 20000 12.4274
half 0 days 00:59:51 21097 13.1091
25k 0 days 01:11:08 25000 15.5343
30k 0 days 01:25:40 30000 18.6411
35k 0 days 01:40:10 35000 21.7480
40k 0 days 01:54:53 40000 24.8548
nettime 0 days 02:01:09 42195 26.2187

Now that we have access to all splits data, we can reshape the data and plot the finish time results with the splits for each mark presented. We will use the horizontal stacked bar charts.

First, let’s fetch the top 10 male and female finishers from Berlin Marathon including their splits.

[210]:

male_top_10 =  man_top_10[['firstname', 'lastname', 'country', '5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime']]
male_top_10['fullname'] = male_top_10['firstname'] + ' ' +  male_top_10['lastname'] + ' (' +  male_top_10['country'] + ')'
male_top_10

[210]:
firstname lastname country 5k 10k 15k 20k half 25k 30k 35k 40k nettime fullname
0 Eliud Kipchoge KEN 0 days 00:14:14 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 00:59:51 0 days 01:11:08 0 days 01:25:40 0 days 01:40:10 0 days 01:54:53 0 days 02:01:09 Eliud Kipchoge (KEN)
1 Mark Korir KEN 0 days 00:14:22 0 days 00:28:56 0 days 00:43:35 0 days 00:58:14 0 days 01:01:26 0 days 01:13:07 0 days 01:28:06 0 days 01:43:25 0 days 01:59:05 0 days 02:05:58 Mark Korir (KEN)
2 Tadu Abate ETH 0 days 00:14:50 0 days 00:29:46 0 days 00:44:40 0 days 00:59:40 0 days 01:02:55 0 days 01:14:44 0 days 01:30:01 0 days 01:44:55 0 days 02:00:03 0 days 02:06:28 Tadu Abate (ETH)
3 Andamlak Belihu ETH 0 days 00:14:16 0 days 00:28:23 0 days 00:42:33 0 days 00:56:45 0 days 00:59:51 0 days 01:11:09 0 days 01:26:11 0 days 01:42:14 0 days 01:59:14 0 days 02:06:40 Andamlak Belihu (ETH)
4 Abel Kipchumba KEN 0 days 00:14:22 0 days 00:28:55 0 days 00:43:35 0 days 00:58:14 0 days 01:01:25 0 days 01:13:07 0 days 01:28:03 0 days 01:43:08 0 days 01:59:14 0 days 02:06:49 Abel Kipchumba (KEN)
5 Limenih Getachew ETH 0 days 00:14:49 0 days 00:29:46 0 days 00:44:40 0 days 00:59:42 0 days 01:02:56 0 days 01:14:45 0 days 01:30:00 0 days 01:45:06 0 days 02:00:22 0 days 02:07:07 Limenih Getachew (ETH)
6 Kenya Sonota JPN 0 days 00:14:49 0 days 00:29:46 0 days 00:44:38 0 days 00:59:39 0 days 01:02:55 0 days 01:14:45 0 days 01:30:01 0 days 01:45:06 0 days 02:00:30 0 days 02:07:14 Kenya Sonota (JPN)
7 Tatsuya Maruyama JPN 0 days 00:15:06 0 days 00:30:10 0 days 00:45:28 0 days 01:00:47 0 days 01:04:05 0 days 01:15:46 0 days 01:30:43 0 days 01:45:53 0 days 02:01:17 0 days 02:07:50 Tatsuya Maruyama (JPN)
8 Kento Kikutani JPN 0 days 00:14:50 0 days 00:29:47 0 days 00:44:39 0 days 00:59:40 0 days 01:02:56 0 days 01:14:45 0 days 01:30:02 0 days 01:45:19 0 days 02:01:10 0 days 02:07:56 Kento Kikutani (JPN)
9 Zablon Chumba KEN 0 days 00:14:49 0 days 00:29:45 0 days 00:44:38 0 days 00:59:39 0 days 01:02:54 0 days 01:14:45 0 days 01:30:00 0 days 01:45:05 0 days 02:00:49 0 days 02:08:01 Zablon Chumba (KEN) 
[211]:

females_top_10 =  female_top_10[['firstname', 'lastname', 'country',  '5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime']]
females_top_10['fullname'] = female_top_10['firstname'] + ' ' +  female_top_10['lastname'] + ' ' + ' (' +  female_top_10['country'] + ')'
females_top_10

[211]:
firstname lastname country 5k 10k 15k 20k half 25k 30k 35k 40k nettime fullname
32 Tigist Assefa ETH 0 days 00:16:22 0 days 00:32:36 0 days 00:48:44 0 days 01:04:43 0 days 01:08:13 0 days 01:20:48 0 days 01:36:41 0 days 01:52:27 0 days 02:08:42 0 days 02:15:37 Tigist Assefa (ETH)
43 Rosemary Wanjiru KEN 0 days 00:16:23 0 days 00:32:38 0 days 00:48:47 0 days 01:04:47 0 days 01:08:17 0 days 01:20:54 0 days 01:36:57 0 days 01:53:16 0 days 02:10:10 0 days 02:18:00 Rosemary Wanjiru (KEN)
44 Tigist Abayechew ETH 0 days 00:16:23 0 days 00:32:38 0 days 00:48:46 0 days 01:04:44 0 days 01:08:14 0 days 01:20:49 0 days 01:36:41 0 days 01:52:46 0 days 02:10:15 0 days 02:18:03 Tigist Abayechew (ETH)
48 Workenesh Edesa ETH 0 days 00:16:22 0 days 00:32:37 0 days 00:48:44 0 days 01:04:44 0 days 01:08:13 0 days 01:20:48 0 days 01:37:01 0 days 01:53:57 0 days 02:11:15 0 days 02:18:51 Workenesh Edesa (ETH)
57 Sisay Meseret Gola ETH 0 days 00:16:23 0 days 00:32:37 0 days 00:48:44 0 days 01:04:43 0 days 01:08:13 0 days 01:20:49 0 days 01:36:42 0 days 01:53:30 0 days 02:12:29 0 days 02:20:58 Sisay Meseret Gola (ETH)
60 Keira D'Amato USA 0 days 00:16:25 0 days 00:32:43 0 days 00:49:11 0 days 01:05:50 0 days 01:09:27 0 days 01:22:36 0 days 01:39:35 0 days 01:57:06 0 days 02:14:28 0 days 02:21:48 Keira D'Amato (USA)
64 Rika Kaseda JPN 0 days 00:16:53 0 days 00:33:30 0 days 00:50:11 0 days 01:06:56 0 days 01:10:33 0 days 01:23:41 0 days 01:40:29 0 days 01:57:27 0 days 02:14:27 0 days 02:21:55 Rika Kaseda (JPN)
65 Ayuko Suzuki JPN 0 days 00:16:53 0 days 00:33:30 0 days 00:50:10 0 days 01:06:55 0 days 01:10:33 0 days 01:23:41 0 days 01:40:30 0 days 01:57:27 0 days 02:14:34 0 days 02:22:02 Ayuko Suzuki (JPN)
66 Sayaka Sato JPN 0 days 00:16:33 0 days 00:33:17 0 days 00:50:03 0 days 01:06:43 0 days 01:10:20 0 days 01:23:29 0 days 01:40:30 0 days 01:57:28 0 days 02:14:48 0 days 02:22:13 Sayaka Sato (JPN)
68 Vibian Chepkirui KEN 0 days 00:16:21 0 days 00:32:36 0 days 00:48:44 0 days 01:04:43 0 days 01:08:13 0 days 01:20:48 0 days 01:36:58 0 days 01:54:12 0 days 02:13:38 0 days 02:22:21 Vibian Chepkirui (KEN) 
[212]:

from matplotlib import pyplot as plt
from matplotlib.pyplot import figure

Now it is time for us to visualize the data! We start with defining the colors of the splits in a dictionary so we can access those later.

[213]:

split_colors = {
    '5k': '#FF3333',
    '10k': '#FFF200',
    '15k': '#EBEBEB',
    '20k': '#39B54A',
    'half': '#00AEEF',
    '25k': '#8c564b',
    '30k': '#e377c2',
    '35k': '#7f7f7f',
    '40k': '#bcbd22',
    'nettime': '#17becf',
}

After that, it is time to (finally) generate the plot!

[214]:



fig, ax = plt.subplots()
fig.set_figheight(10)
fig.set_figwidth(15)

SPLITS = ['5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime']

for _, finisher in male_top_10.iterrows():
    previous_split_end = pd.to_timedelta("0:00:00").total_seconds()
    for split in SPLITS :
        ax.barh(
            [finisher['fullname']],
            finisher[split].total_seconds() - previous_split_end,
            left=previous_split_end,
            color=split_colors[split],
            label = split,
            edgecolor = "black"
        )
        previous_split_end = finisher[split].total_seconds()

def timeTicks(x, pos):
    d = datetime.timedelta(seconds=x)
    return str(d)
formatter = matplotlib.ticker.FuncFormatter(timeTicks)
ax.xaxis.set_major_formatter(formatter)
ax.xaxis.set_major_locator(plt.MultipleLocator(15*30))

# Set title
plt.title(f'Race splits - Berlin Marathon 2022')

# Set x-label
plt.xlabel('Time')

# Invert y-axis
plt.gca().invert_yaxis()

# Remove frame from plot
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['left'].set_visible(False)


ax.set_axisbelow(True)
ax.xaxis.grid(True, which='major', linestyle='--', color='black', zorder=-1000)

horiz_offset = 1.03
vert_offset = 1.
ax.legend(SPLITS, bbox_to_anchor=(horiz_offset, vert_offset))
plt.show()
../_images/user_guide_5-marathon_analysis_38_0.png

We can see above that the Eliud Kipchoge, since the second split (10k) took the leadership with some minutes at front of the other runners. Let’s now plot the race splits for the top ten female finishers.

[215]:



fig, ax = plt.subplots()
fig.set_figheight(10)
fig.set_figwidth(15)

SPLITS = ['5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime']

for _, finisher in females_top_10.iterrows():
    previous_split_end = pd.to_timedelta("0:00:00").total_seconds()
    for split in SPLITS :
        ax.barh(
            [finisher['fullname']],
            finisher[split].total_seconds() - previous_split_end,
            left=previous_split_end,
            color=split_colors[split],
            label = split,
            edgecolor = "black"
        )
        previous_split_end = finisher[split].total_seconds()

def timeTicks(x, pos):
    d = datetime.timedelta(seconds=x)
    return str(d)
formatter = matplotlib.ticker.FuncFormatter(timeTicks)
ax.xaxis.set_major_formatter(formatter)
ax.xaxis.set_major_locator(plt.MultipleLocator(15*30))

# Set title
plt.title(f'Race splits - Berlin Marathon 2022')

# Set x-label
plt.xlabel('Time')

# Invert y-axis
plt.gca().invert_yaxis()

# Remove frame from plot
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
ax.spines['left'].set_visible(False)


ax.set_axisbelow(True)
ax.xaxis.grid(True, which='major', linestyle='--', color='black', zorder=-1000)

horiz_offset = 1.03
vert_offset = 1.
ax.legend(SPLITS, bbox_to_anchor=(horiz_offset, vert_offset))
plt.show()
../_images/user_guide_5-marathon_analysis_40_0.png

Different from the female finishers, where the finish podium was more disputed. Tigist Assefa only took the leadership with relative difference from the split 40k (almost at the end of the race).

One of the curiosities about the race is to visualize the standing changes over the splits. To see if any runners lost their positions during the first or the second half of the race, or if any runners lost the podium during the final splits. Let’s prepate the data for the plot, by calculating the runner’s position at each split of the race.

We will create a rank table based on the split times for each runner. Pandas comes with a useful method for this task named pandas.DataFrame.rank. We will compute for each split the standing position, so we can have a overall look at the changes during the race.

[216]:

male_top_10_filtered = male_top_10.drop(['firstname', 'lastname', 'country'], axis=1)
ranked_male_top10 = male_top_10_filtered.set_index('fullname').rank(method='dense').reset_index()
ranked_male_top10

[216]:
fullname 5k 10k 15k 20k half 25k 30k 35k 40k nettime
0 Eliud Kipchoge (KEN) 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
1 Mark Korir (KEN) 3.0 3.0 2.0 2.0 3.0 3.0 4.0 4.0 2.0 2.0
2 Tadu Abate (ETH) 5.0 5.0 5.0 4.0 5.0 4.0 6.0 5.0 4.0 3.0
3 Andamlak Belihu (ETH) 2.0 1.0 1.0 1.0 1.0 2.0 2.0 2.0 3.0 4.0
4 Abel Kipchumba (KEN) 3.0 2.0 2.0 2.0 2.0 3.0 3.0 3.0 3.0 5.0
5 Limenih Getachew (ETH) 4.0 5.0 5.0 5.0 6.0 5.0 5.0 7.0 5.0 6.0
6 Kenya Sonota (JPN) 4.0 5.0 3.0 3.0 5.0 5.0 6.0 7.0 6.0 7.0
7 Tatsuya Maruyama (JPN) 6.0 7.0 6.0 6.0 7.0 6.0 8.0 9.0 9.0 8.0
8 Kento Kikutani (JPN) 5.0 6.0 4.0 4.0 6.0 5.0 7.0 8.0 8.0 9.0
9 Zablon Chumba (KEN) 4.0 4.0 3.0 3.0 4.0 5.0 5.0 6.0 7.0 10.0

Now, we’re melting the dataset based on the split. This will convert the dataset from wide to long, were multiple columns (in this case the split and the runner) will work as identifiers. The result of the melting looks like this:

[217]:

berlin_male_top10_standings = pd.melt(ranked_male_top10, ['fullname'])
berlin_male_top10_standings

[217]:
fullname variable value
0 Eliud Kipchoge (KEN) 5k 1.0
1 Mark Korir (KEN) 5k 3.0
2 Tadu Abate (ETH) 5k 5.0
3 Andamlak Belihu (ETH) 5k 2.0
4 Abel Kipchumba (KEN) 5k 3.0
... ... ... ...
95 Limenih Getachew (ETH) nettime 6.0
96 Kenya Sonota (JPN) nettime 7.0
97 Tatsuya Maruyama (JPN) nettime 8.0
98 Kento Kikutani (JPN) nettime 9.0
99 Zablon Chumba (KEN) nettime 10.0

100 rows × 3 columns

Now we have the data in the format we want it in, we can generate the final plot.

[218]:

import seaborn as sns
import matplotlib.pyplot as plt

[219]:

# Increase the size of the plot
sns.set(rc={'figure.figsize':(11.7,8.27)})

# Initiate the plot
fig, ax = plt.subplots()

# Set the title of the plot
ax.set_title("Berlin Marathon 2022 - Male")

# Draw a line for every runner in the data by looping through all the standings
for runner in pd.unique(berlin_male_top10_standings['fullname']):
    sns.lineplot(
        x='variable',
        y='value',
        data=berlin_male_top10_standings.loc[berlin_male_top10_standings['fullname']==runner],
    )

# Invert Y-axis to have runner leader (#1) on top
ax.invert_yaxis()

# Set the values that appear on  y-axes
ax.set_yticks(range(1, 10))

# Set the labels of the axes
ax.set_xlabel("Split")
ax.set_ylabel("Race position")

# Disable the gridlines
ax.grid(False)


# Add the runner name to the lines
for line, name in zip(ax.lines, ranked_male_top10['fullname'].tolist()):
    y = line.get_ydata()[-1]
    x = line.get_xdata()[-1]

    text = ax.annotate(
        name,
        xy=(x + 0.1, y),
        xytext=(0, 0),
        color=line.get_color(),
        xycoords=(
            ax.get_xaxis_transform(),
            ax.get_yaxis_transform()
        ),
        textcoords="offset points"
    )
../_images/user_guide_5-marathon_analysis_49_0.png

Amazing to see the performance of Eliud Kipchoge, always on the lead. Other runner that lost positions was Andamlak Belihu (ETH), who tried to follow Eliud but from the second half lost three positions.

Let’s analyze the women’s standing during the race. Let’s perform the transformation on the female’s dataset (ranking and melting methods).

[220]:

females_top_10_filtered = females_top_10.drop(['firstname', 'lastname', 'country'], axis=1)
ranked_females_top10 = females_top_10_filtered.set_index('fullname').rank(method='dense').reset_index()
ranked_females_top10

[220]:
fullname 5k 10k 15k 20k half 25k 30k 35k 40k nettime
0 Tigist Assefa (ETH) 2.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0
1 Rosemary Wanjiru (KEN) 3.0 3.0 3.0 3.0 3.0 3.0 3.0 3.0 2.0 2.0
2 Tigist Abayechew (ETH) 3.0 3.0 2.0 2.0 2.0 2.0 1.0 2.0 3.0 3.0
3 Workenesh Edesa (ETH) 2.0 2.0 1.0 2.0 1.0 1.0 5.0 5.0 4.0 4.0
4 Sisay Meseret Gola (ETH) 3.0 2.0 1.0 1.0 1.0 2.0 2.0 4.0 5.0 5.0
5 Keira D'Amato (USA) 4.0 4.0 4.0 4.0 4.0 4.0 6.0 7.0 8.0 6.0
6 Rika Kaseda (JPN) 6.0 6.0 7.0 7.0 6.0 6.0 7.0 8.0 7.0 7.0
7 Ayuko Suzuki (JPN) 6.0 6.0 6.0 6.0 6.0 6.0 8.0 8.0 9.0 8.0
8 Sayaka Sato (JPN) 5.0 5.0 5.0 5.0 5.0 5.0 8.0 9.0 10.0 9.0
9 Vibian Chepkirui (KEN) 1.0 1.0 1.0 1.0 1.0 1.0 4.0 6.0 6.0 10.0 
[221]:

berlin_female_top10_standings = pd.melt(ranked_females_top10, ['fullname'])
berlin_female_top10_standings

[221]:
fullname variable value
0 Tigist Assefa (ETH) 5k 2.0
1 Rosemary Wanjiru (KEN) 5k 3.0
2 Tigist Abayechew (ETH) 5k 3.0
3 Workenesh Edesa (ETH) 5k 2.0
4 Sisay Meseret Gola (ETH) 5k 3.0
... ... ... ...
95 Keira D'Amato (USA) nettime 6.0
96 Rika Kaseda (JPN) nettime 7.0
97 Ayuko Suzuki (JPN) nettime 8.0
98 Sayaka Sato (JPN) nettime 9.0
99 Vibian Chepkirui (KEN) nettime 10.0

100 rows × 3 columns

Let’s now plot the standings position through the race.

[222]:

# Increase the size of the plot
sns.set(rc={'figure.figsize':(11.7,8.27)})

# Initiate the plot
fig, ax = plt.subplots()

# Set the title of the plot
ax.set_title("Berlin Marathon 2022 - Female")

# Draw a line for every runner in the data by looping through all the standings
for runner in pd.unique(berlin_female_top10_standings['fullname']):
    sns.lineplot(
        x='variable',
        y='value',
        data=berlin_female_top10_standings.loc[berlin_female_top10_standings['fullname']==runner],
    )

# Invert Y-axis to have runner leader (#1) on top
ax.invert_yaxis()

# Set the values that appear on  y-axes
ax.set_yticks(range(1, 10))

# Set the labels of the axes
ax.set_xlabel("Split")
ax.set_ylabel("Race position")

# Disable the gridlines
ax.grid(False)


# Add the runner name to the lines
for line, name in zip(ax.lines, ranked_females_top10['fullname'].tolist()):
    y = line.get_ydata()[-1]
    x = line.get_xdata()[-1]

    text = ax.annotate(
        name,
        xy=(x + 0.1, y),
        xytext=(0, 0),
        color=line.get_color(),
        xycoords=(
            ax.get_xaxis_transform(),
            ax.get_yaxis_transform()
        ),
        textcoords="offset points"
    )
../_images/user_guide_5-marathon_analysis_55_0.png

Tigist Assefa (ETH) showing a steadily high performance maintaning the leadership during all the race. It is interesting to see the that Vibian Chepkirui (KEN) tried to follow her until the split 25k and lost many position, finishing in 10th place. The same happened to Workensh Edesa (ETH) alternating the first and second positions until the second half of the race, which she lost three positions and finished on 4th place.

Analyzing Eliud Kipchoge results#

Eliud Kipchoge shattered his own world record on Berlin Marathon with a time of 02:01:09. Kipchoge’s previous best in an official 42.2km race was 2:01:39 set on the same course in 2018. Eliud also made history in 2019 when he became the first man to run a marathon in under two hours at INEOS 1:59 Challenge, in Vienna, Austria. Based on these amazing finishing times, we would like to compare all the split datas from these races to see how Eliud performed during the race compared to his previous world record in 2018 and INEOS 1:59 Challenge, so we can see what he need to perform this mark in a marathon again.

First we need to collect his available splits in 2022 and 2018 Berlin Marathons and the INEOS 1:49 Challenge.

[223]:

#collecting the splits from 2022 from the previous berlin dataset
eliud_2022_berlin_result = race_result.splits.pick_athlete(identifier='1')
eliud_2022_berlin_result.rename(columns={'time': 'time_2022'}, inplace=True)
eliud_2022_berlin_result

[223]:
time_2022 distance_meters distance_miles
split
0k 0 days 00:00:00 0 0.0000
5k 0 days 00:14:14 5000 3.1069
10k 0 days 00:28:23 10000 6.2137
15k 0 days 00:42:33 15000 9.3206
20k 0 days 00:56:45 20000 12.4274
half 0 days 00:59:51 21097 13.1091
25k 0 days 01:11:08 25000 15.5343
30k 0 days 01:25:40 30000 18.6411
35k 0 days 01:40:10 35000 21.7480
40k 0 days 01:54:53 40000 24.8548
nettime 0 days 02:01:09 42195 26.2187 
[224]:

#We created the splits pandas Series from the 2018 Berlin Marathon for Eliud Kipchoge manually.  Source: Berlin Marathon Result Archive
eliud_2018_berlin_result = pd.Series(["0:00:00", "00:14:24", "00:29:01", "00:43:38", "00:57:56", "01:01:06", "01:12:24", "01:26:45", "01:41:03", "01:55:32", "02:01:39"], index=['0k', '5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime'], name='time_2018')
eliud_2018_berlin_result = pd.to_timedelta(eliud_2018_berlin_result)

eliud_2018_berlin_result

[224]:

0k        0 days 00:00:00
5k        0 days 00:14:24
10k       0 days 00:29:01
15k       0 days 00:43:38
20k       0 days 00:57:56
half      0 days 01:01:06
25k       0 days 01:12:24
30k       0 days 01:26:45
35k       0 days 01:41:03
40k       0 days 01:55:32
nettime   0 days 02:01:39
Name: time_2018, dtype: timedelta64[ns]

[225]:

#We created the splits pandas Series from the 2019 INEOS 159 Challenge for Eliud Kipchoge manually. Source: wikipedia
eliud_2019_ineos_159_result = pd.Series(["0:00:00", "00:14:10", "00:28:20", "00:42:34", "00:56:47", "0:59:37", "01:10:59", "01:25:11", "01:39:23", "01:53:36", "01:59:40"], index=['0k', '5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime'], name='time_ineos_159')
eliud_2019_ineos_159_result = pd.to_timedelta(eliud_2019_ineos_159_result)

eliud_2019_ineos_159_result

[225]:

0k        0 days 00:00:00
5k        0 days 00:14:10
10k       0 days 00:28:20
15k       0 days 00:42:34
20k       0 days 00:56:47
half      0 days 00:59:37
25k       0 days 01:10:59
30k       0 days 01:25:11
35k       0 days 01:39:23
40k       0 days 01:53:36
nettime   0 days 01:59:40
Name: time_ineos_159, dtype: timedelta64[ns]

[226]:

#Merging all series into a single dataframe containing all splits from 2018, Ineos 159 and 2022 race results.
eliud_race_results = eliud_2022_berlin_result.merge(eliud_2019_ineos_159_result.to_frame(), left_index=True, right_index=True).merge(eliud_2018_berlin_result.to_frame(), left_index=True, right_index=True)
eliud_race_results

[226]:
time_2022 distance_meters distance_miles time_ineos_159 time_2018
0k 0 days 00:00:00 0 0.0000 0 days 00:00:00 0 days 00:00:00
5k 0 days 00:14:14 5000 3.1069 0 days 00:14:10 0 days 00:14:24
10k 0 days 00:28:23 10000 6.2137 0 days 00:28:20 0 days 00:29:01
15k 0 days 00:42:33 15000 9.3206 0 days 00:42:34 0 days 00:43:38
20k 0 days 00:56:45 20000 12.4274 0 days 00:56:47 0 days 00:57:56
half 0 days 00:59:51 21097 13.1091 0 days 00:59:37 0 days 01:01:06
25k 0 days 01:11:08 25000 15.5343 0 days 01:10:59 0 days 01:12:24
30k 0 days 01:25:40 30000 18.6411 0 days 01:25:11 0 days 01:26:45
35k 0 days 01:40:10 35000 21.7480 0 days 01:39:23 0 days 01:41:03
40k 0 days 01:54:53 40000 24.8548 0 days 01:53:36 0 days 01:55:32
nettime 0 days 02:01:09 42195 26.2187 0 days 01:59:40 0 days 02:01:39

Let’s now replay these four three WRs in the same (virtual) race, to get a better sense of how Eliud Kipchoge has performed by comparing their pacing and timing information.

[227]:

#let's compute the timing difference for all the splits againts the INEOS 159 WR record (best result from Eliud)
eliud_race_results['diff_2022_ineos'] = eliud_race_results['time_2022'].dt.total_seconds() - eliud_race_results['time_ineos_159'].dt.total_seconds()
eliud_race_results['diff_2018_ineos'] = eliud_race_results['time_2018'].dt.total_seconds() - eliud_race_results['time_ineos_159'].dt.total_seconds()
eliud_race_results['ineos_159_time_ref'] = 0.0 #all values 0 as reference
eliud_race_results['diff_2022_2018'] = eliud_race_results['time_2022'].dt.total_seconds() - eliud_race_results['time_2018'].dt.total_seconds()
eliud_race_results

[227]:
time_2022 distance_meters distance_miles time_ineos_159 time_2018 diff_2022_ineos diff_2018_ineos ineos_159_time_ref diff_2022_2018
0k 0 days 00:00:00 0 0.0000 0 days 00:00:00 0 days 00:00:00 0.0 0.0 0.0 0.0
5k 0 days 00:14:14 5000 3.1069 0 days 00:14:10 0 days 00:14:24 4.0 14.0 0.0 -10.0
10k 0 days 00:28:23 10000 6.2137 0 days 00:28:20 0 days 00:29:01 3.0 41.0 0.0 -38.0
15k 0 days 00:42:33 15000 9.3206 0 days 00:42:34 0 days 00:43:38 -1.0 64.0 0.0 -65.0
20k 0 days 00:56:45 20000 12.4274 0 days 00:56:47 0 days 00:57:56 -2.0 69.0 0.0 -71.0
half 0 days 00:59:51 21097 13.1091 0 days 00:59:37 0 days 01:01:06 14.0 89.0 0.0 -75.0
25k 0 days 01:11:08 25000 15.5343 0 days 01:10:59 0 days 01:12:24 9.0 85.0 0.0 -76.0
30k 0 days 01:25:40 30000 18.6411 0 days 01:25:11 0 days 01:26:45 29.0 94.0 0.0 -65.0
35k 0 days 01:40:10 35000 21.7480 0 days 01:39:23 0 days 01:41:03 47.0 100.0 0.0 -53.0
40k 0 days 01:54:53 40000 24.8548 0 days 01:53:36 0 days 01:55:32 77.0 116.0 0.0 -39.0
nettime 0 days 02:01:09 42195 26.2187 0 days 01:59:40 0 days 02:01:39 89.0 119.0 0.0 -30.0 
[228]:

fig, ax1 = plt.subplots(figsize=(12,4))
ax1.plot(eliud_race_results.index, eliud_race_results['diff_2022_ineos'], color="Slateblue",
         alpha=0.6, linewidth=2,  linestyle='dashed', marker='o', label='Berlin Marathon 2022')
ax1.plot(eliud_race_results.index,  eliud_race_results['diff_2018_ineos'], marker='o', linestyle='dashed', label='Berlin Marathon 2018' )
ax1.plot(eliud_race_results.index,  eliud_race_results['ineos_159_time_ref'], marker='o', linestyle='dashed', label='INEOS 159 Challenge' )
ax1.set_xlabel('Race Segments(km)', size=12)
ax1.set_ylabel('Seconds behind INEOS 159 WR (2019)', size=12)
ax1.grid()
ax1.legend(loc=0)
plt.show()
../_images/user_guide_5-marathon_analysis_66_0.png

Using Kipchoge’s INEO Challenge run as the baseline, Figure above shows the number of seconds each WR run was behind Kipchoge best result at INEOS 159 challenge at the end of each race segment (and at the finish-line) in this virtual race. Kipchoge 2022’s result showed an amazing start where by 15k-20km marks , he has ahead from his best result from INEOS for about two seconds, but gradually extends his difference time after the half mark. We can also see that the 2022 vss 2018 previous record, what an amazing difference, finishing with 30 seconds ahead. By any objective measure Kipchoge’s Berlin world-record 2022 race was nothing short of stunning. He obliterated his own previous 2018 record.

[229]:

import numpy as np
#compute the distance ellapsed per segment , and convert it to kms
dist_diff = (eliud_race_results["distance_meters"] /1000).diff().fillna(eliud_race_results['distance_meters'][0])

for timing_race, pace_race in [('time_2022', 'pace_2022'), ('time_ineos_159', 'pace_ineos_159' ), ('time_2018', 'pace_2018')]:
    #compute the time ellapsed per segment
    time_diff = eliud_race_results[timing_race].diff().fillna(eliud_race_results[timing_race][0])  / np.timedelta64(1, "s")
    #compute the speed by dividing the split distance and the split time ellapsed and convert it back to timedelta
    speed = dist_diff / time_diff
    pace_timedelta = pd.to_timedelta(1 / speed, unit="s")
    eliud_race_results[pace_race] = pace_timedelta
eliud_race_results

[229]:
time_2022 distance_meters distance_miles time_ineos_159 time_2018 diff_2022_ineos diff_2018_ineos ineos_159_time_ref diff_2022_2018 pace_2022 pace_ineos_159 pace_2018
0k 0 days 00:00:00 0 0.0000 0 days 00:00:00 0 days 00:00:00 0.0 0.0 0.0 0.0 NaT NaT NaT
5k 0 days 00:14:14 5000 3.1069 0 days 00:14:10 0 days 00:14:24 4.0 14.0 0.0 -10.0 0 days 00:02:50.800000 0 days 00:02:50 0 days 00:02:52.800000
10k 0 days 00:28:23 10000 6.2137 0 days 00:28:20 0 days 00:29:01 3.0 41.0 0.0 -38.0 0 days 00:02:49.800000 0 days 00:02:50 0 days 00:02:55.400000
15k 0 days 00:42:33 15000 9.3206 0 days 00:42:34 0 days 00:43:38 -1.0 64.0 0.0 -65.0 0 days 00:02:50 0 days 00:02:50.800000 0 days 00:02:55.400000
20k 0 days 00:56:45 20000 12.4274 0 days 00:56:47 0 days 00:57:56 -2.0 69.0 0.0 -71.0 0 days 00:02:50.400000 0 days 00:02:50.600000 0 days 00:02:51.600000
half 0 days 00:59:51 21097 13.1091 0 days 00:59:37 0 days 01:01:06 14.0 89.0 0.0 -75.0 0 days 00:02:49.553327256 0 days 00:02:34.968094804 0 days 00:02:53.199635369
25k 0 days 01:11:08 25000 15.5343 0 days 01:10:59 0 days 01:12:24 9.0 85.0 0.0 -76.0 0 days 00:02:53.456315655 0 days 00:02:54.737381501 0 days 00:02:53.712528824
30k 0 days 01:25:40 30000 18.6411 0 days 01:25:11 0 days 01:26:45 29.0 94.0 0.0 -65.0 0 days 00:02:54.400000 0 days 00:02:50.400000 0 days 00:02:52.200000
35k 0 days 01:40:10 35000 21.7480 0 days 01:39:23 0 days 01:41:03 47.0 100.0 0.0 -53.0 0 days 00:02:54 0 days 00:02:50.400000 0 days 00:02:51.600000
40k 0 days 01:54:53 40000 24.8548 0 days 01:53:36 0 days 01:55:32 77.0 116.0 0.0 -39.0 0 days 00:02:56.600000 0 days 00:02:50.600000 0 days 00:02:53.800000
nettime 0 days 02:01:09 42195 26.2187 0 days 01:59:40 0 days 02:01:39 89.0 119.0 0.0 -30.0 0 days 00:02:51.298405467 0 days 00:02:45.831435080 0 days 00:02:47.198177677 
[230]:

#remove the 0k from the frame, since it doesn't give us any useful information.
eliud_race_results_filtered = eliud_race_results.iloc[1:]

#Data for visualization
splits = eliud_race_results_filtered.index.tolist()
split_paces = {}
mean_paces = {}
for pace_race in ['pace_2022', 'pace_ineos_159', 'pace_2018']:
    #convert the timedelta paces to float and convert it into a list.
    split_paces[pace_race] = (eliud_race_results_filtered[pace_race] / datetime.timedelta(minutes=1)).tolist()
    #compute the mean pace and convert into a float. Repeats it by the number of splits.
    mean_paces[pace_race] = [(eliud_race_results_filtered[pace_race].mean()) / datetime.timedelta(minutes=1)] * 10
splits

[230]:

['5k', '10k', '15k', '20k', 'half', '25k', '30k', '35k', '40k', 'nettime']

[231]:

#Obtain Angles for the radar plot
angles=np.linspace(0,2*np.pi,len(splits), endpoint=False)
angles

[231]:

array([0.        , 0.62831853, 1.25663706, 1.88495559, 2.51327412,
       3.14159265, 3.76991118, 4.39822972, 5.02654825, 5.65486678])

[232]:

#Completing full circle

angles=np.concatenate((angles,[angles[0]]))
splits.append(splits[0])
for pace_race in split_paces:
    split_paces[pace_race].append(split_paces[pace_race][0])
    mean_paces[pace_race].append(mean_paces[pace_race][0])

[233]:

#Simple chart for one race
plt.style.use('ggplot')

fig=plt.figure(figsize=(6,6))
ax=fig.add_subplot(polar=True)
#basic plot
ax.plot(angles,split_paces['pace_2022'], 'o--', color='g', label='Berlin Marathon 2022')
#fill plot
ax.fill(angles, split_paces['pace_2022'], alpha=0.25, color='g')
#Add labels
ax.set_thetagrids(angles * 180/np.pi, splits)
ax.set_ylim(2.4, 3.0)

ax.set_yticklabels([str(datetime.timedelta(minutes=x)) if idx % 2 != 0 else '' for idx, x in  enumerate(ax.get_yticks()) ], rotation = 45, zorder = 500)
ax.grid(color='#AAAAAA')
plt.grid(True)
plt.tight_layout()
plt.legend()
plt.show()
../_images/user_guide_5-marathon_analysis_72_0.png 
[250]:

#Let's combine all the races in the same chart now

fig=plt.figure(figsize=(12,12))
ax1=fig.add_subplot(121, polar=True)
ax2=fig.add_subplot(122, polar=True)

#2022 Berlin Plot
ax1.plot(angles,split_paces['pace_2022'], 'o--', color='#1aaf6c', label='Berlin Marathon 2022')
ax1.fill(angles, split_paces['pace_2022'], alpha=0.25, color='#1aaf6c')

ax2.plot(angles,mean_paces['pace_2022'], 'o--', color='#1aaf6c', label='Berlin Marathon 2022')
ax2.fill(angles, mean_paces['pace_2022'], alpha=0.25, color='#1aaf6c')

#2018 Berlin Plot
ax1.plot(angles,split_paces['pace_2018'], 'o-', color='#429bf4', linewidth=1, label='Berlin Marathon 2018')
ax1.fill(angles, split_paces['pace_2018'], alpha=0.25, color='#429bf4')

ax2.plot(angles,mean_paces['pace_2018'], 'o-', color='#429bf4', label='Berlin Marathon 2018')
ax2.fill(angles, mean_paces['pace_2018'], alpha=0.25, color='#429bf4')

#Ineos 159 Plot
ax1.plot(angles,split_paces['pace_ineos_159'], 'o-', color='#d42cea', linewidth=1, label='INEOS 159')
ax1.fill(angles, split_paces['pace_ineos_159'], alpha=0.25, color='#d42cea')

ax2.plot(angles,mean_paces['pace_ineos_159'], 'o-', color='#d42cea', label='INEOS 159')
ax2.fill(angles, mean_paces['pace_ineos_159'], alpha=0.25, color='#d42cea')

ax1.set_thetagrids(angles * 180/np.pi, splits)
ax1.set_ylim(2.4, 3.0)
ax2.set_thetagrids(angles * 180/np.pi, splits)
ax2.set_ylim(2.4, 3.0)


ax1.set_yticklabels([str(datetime.timedelta(minutes=x)) if idx % 2 != 0 else '' for idx, x in  enumerate(ax1.get_yticks()) ], rotation = 45, zorder = 500)
ax1.grid(color='#AAAAAA')

ax2.set_yticklabels([str(datetime.timedelta(minutes=x)) if idx % 2 != 0 else '' for idx, x in  enumerate(ax2.get_yticks()) ], rotation = 45, zorder = 500)
ax2.grid(color='#AAAAAA')


ax1.set_title('Comparing Eliud Kipchogue Pace Splits (min/km) ', y=1.08)
ax2.set_title('Comparing Eliud Kipchogue Mean Pace Splits (min/km) ', y=1.08)

ax1.grid(True)
ax2.grid(True)
plt.tight_layout()
plt.legend(loc='upper right', bbox_to_anchor=(1.3, 1.1))

plt.show()
../_images/user_guide_5-marathon_analysis_73_0.png

The figure at the left above shows the pacing (in mins/km) for all 3 WRs across each of the segments of the race (5 km, 10 km, …, 40km), and the final 2.195km segment. The second figure at right, the plot reflects the average pace for the races evaluated.

For those not familiar with this type of chart, we call it radar chart, also called as Spider chart or Web chart. It is a graphical method used for comparing multiple quantitative variables. It is a two dimensional polar visualization.

Based on the timing data released by the Berlin Marathon 2022, Kipchoge ran the first 5 km in 14 mins, 14 seconds, or about 2:50 mins/km. Between 5km and 20km he ran between 2:49 min/km and 2:50 min/km, the same pace he did at 159 INEOS challenge. but after the 20km checkpoint he started to slow the pace (the final half of the race), even lower than the 2018 WR Berlin Marathon, but comfortable enough to finish with the WR. Imagine if he could keep the same pace for the second part of the race, it would be amazing to see him at a regular marathon to finish under 2hr mark.

Conclusions#

By any objective measure Kipchoge’s Berlin world-record 2022 race is considered amazing, setting on the same course, his second record. It is impressive also his performance before the halfway point, with his incredibly pace similar to 159 INEOS challenge. It seems that Kipchoge is the fatest marathoner and we will might see him showing to the world that no human is limited!