Data for on players in Pro Ladder is released on playgwent.com, but it is limited. You get the rank, the score, the country of origin and the number of matches played. Using some fairly basic data analysis tricks there must be more we can do with these data! Using python we’ll scrape data for the top Gwent players on pro-ladder and calculate additional statistics about the current season, popularity of the game in different countries, players’ efficiency, players national rank, …

A jupyter notebook with all code can be found on GitHub which you can explore through Binder without installing anything. For those that want to check their own national rank or ladder efficiency index (and don’t care about the code). Download links for the full tables discussed here (in Excel-format) are available here:

  • Player Statistics : Player data for each season, with ladder efficiency and national rank added.
  • Player Summaries : Summary data for each player that made an appearance in pro rank. Includes number of appearances on the leaderboards, min and max MMR, best rank, best national rank, …
  • Seasonal summary : Number of games played each season in Masters 2, minimum and maximum MMR as well as top 500,200 and 64 cutoffs.
  • National Statistics : Data per country, number of pro players per million inhabitants, …

Update 03/11/2020: All downloadable files were updated and now contain data from the last season (Season of the Cat).

Update 03/09/2020: Credit where credit is due! After putting this blog post up I found two articles by Lerio2 that predate mine where he did the same analysis to check the popularity and rank countries (based on teams of 4 players). Though I did my analysis independently, he had the idea several months earlier and deserves full credit for that! You can read his articles, called Nations of Gwent, here and here

Getting the Data

Python has two powerful packages to scrape data from the web: the requests library to download data and BeautifulSoup to parse the HTML that comes back and extract information. The tabular data from playgwent.com is pretty straightforward to parse. There is the rank, the player’s handle, the number of matches played and their score (which is called the Matchmaking Rating or MMR).

Furthermore, there is a flag icon indicating the country the player is from. These icons have a class that contains the two letter code, which follows the official ISO 3166 international standard.

<i class="flag-icon flag-icon-pl"></i>

The two letter code can be easily extracted from the the html tag, while converting it to a human readable name can be done in a few lines of code using the python library pycountry. As shown in the stub below, you can provide it with a two letter code (pl in the example below) and it will return all other names, including the common name (Poland here). So after scraping the data, the pycountry library was used to get proper names for all countries.

import pycountry
pycountry.countries.get(alpha_2='pl')

# Output: 
# Country(alpha_2='PL', alpha_3='POL', name='Poland', numeric='616', official_name='Republic of Poland')

While reading the data we’ll also keep track of which players were in the top 500 the season before (note that this does require all seasons to be loaded and in orde). So we end up with a table (called full_df in the code), that looks like this:

rank name country matches mmr season previous_top500
1 kolemoen Germany 431 10484 M2_01 Wolf 2020 no
2 kams134 Poland 923 10477 M2_01 Wolf 2020 no
3 TailBot Poland 538 10472 M2_01 Wolf 2020 no
4 Pajabol Poland 820 10471 M2_01 Wolf 2020 no
5 Adzikov Poland 1105 10442 M2_01 Wolf 2020 no

Adding National Rank and Efficiency Statistics

The rank on playgwent.com is the global rank, adding a national rank can be done in a single line of code. The groupby function in combination with the rank function does exactly what we want here.

full_df['national_rank'] = full_df.groupby(['country','season'])["mmr"].rank("first", ascending=False)

In Gwent you need to play at least 25 games with four out of six factions. This will give you a base score, MMR, of 9600. Winning a game increases the MMR, depending on the current rank of your opponent (usually about 7 points are gained) and losing costs you MMR points. The highest reached MMR per faction is summed up to get the final score. So with a higher win-rate, better scores can be obtained with fewer games. To find out which players are more efficient in climbing (and arguably better at the game than others at the same MMR) we we take the MMR, subtract the base value (9600) and divide by the number of matches. However, as increasing the MMR score becomes progressively more difficult as players will face better opponents as they climb the ladder, Lerio2 from Team Legacy proposed to divide by the square root of the number of matches. Their metric, the Ladder Efficiency Index or LEI is calculated here as well.

full_df['efficiency'] = ((full_df['mmr']-9600))/full_df['matches']
full_df['lei'] = ((full_df['mmr']-9600))/np.sqrt(full_df['matches'])

Now our full dataframe has two additional columns one with the simple linear efficiency and one with Team Legacy’s Ladder Efficiency Index.

rank name country matches mmr season previous_top500 national_rank efficiency lei
1 kolemoen Germany 431 10484 M2_01 Wolf 2020 no 1.0 2.051044 42.580782
2 kams134 Poland 923 10477 M2_01 Wolf 2020 no 1.0 0.950163 28.866807
3 TailBot Poland 538 10472 M2_01 Wolf 2020 no 2.0 1.620818 37.594590
4 Pajabol Poland 820 10471 M2_01 Wolf 2020 no 3.0 1.062195 30.416639
5 Adzikov Poland 1105 10442 M2_01 Wolf 2020 no 4.0 0.761991 25.329753

You can download the full table here.

Season Summary

About every month or so there is a new season in Gwent. Using the groupby function we can very quickly create a summary how many games were played by the pro-ranked players (do note that only the 2860 best players are listed on the website). We’ll also add the cutoff values for rank 500, 200 and 64 as these are important thresholds for competitive players. Here the aggregate function agg is used in combination with NamedAgg to calculate all statistics in one go.

per_season_df = full_df.groupby(['season']).agg(
    min_mmr     = pd.NamedAgg('mmr', 'min'),
    max_mmr     = pd.NamedAgg('mmr', 'max'),
    num_matches = pd.NamedAgg('matches', 'sum')
).reset_index()

top500_cutoffs = full_df[full_df['rank'] == 500][['season', 'mmr']].rename(columns={'mmr': 'top500_cutoff'})
top200_cutoffs = full_df[full_df['rank'] == 200][['season', 'mmr']].rename(columns={'mmr': 'top200_cutoff'})
top64_cutoffs  = full_df[full_df['rank'] == 64][['season', 'mmr']].rename(columns={'mmr': 'top64_cutoff'})

per_season_df = pd.merge(per_season_df, top500_cutoffs, on='season')
per_season_df = pd.merge(per_season_df, top200_cutoffs, on='season')
per_season_df = pd.merge(per_season_df, top64_cutoffs, on='season')
per_season_df

The full output from this you can see below:

season min_mmr max_mmr num_matches top500_cutoff top200_cutoff top64_cutoff
M2_01 Wolf 2020 2407 10484 699496 9749 9872 10061
M2_02 Love 2020 7776 10537 769358 9832 9952 10117
M2_03 Bear 2020 9427 10669 862678 9867 9995 10204
M2_04 Elf 2020 9666 10751 1004830 9952 10087 10293
M2_05 Viper 2020 9635 10622 859640 9910 10028 10255
M2_06 Magic 2020 9624 10597 793401 9896 10002 10191
M2_07 Griffin 2020 9698 10667 996742 9978 10100 10289
M2_08 Draconid 2020 9666 10546 838212 9946 10061 10246

The number of matches played by the top players is an indication how many people are playing the game, as more active players would require more games to be played to climb pro ladder. You can see that the popularity peaked during the Season of the Elves. During this season also some new leader abilities were introduced, so the fresh content could also to players return to the game. A similar increase in matches can be seen in the Season of the Griffin with the release of new cards through the Master Mirror expansion. So it seems that new content is a good incentive for players to play more, and spark a fiercer competition.

You can download the full table here.

Where is Gwent Being Played

So using the groupby function in combination with the agg we can very quickly count how many pro players there are per country. We can then combine this with the population size of each country (and somewhat up-to-date list can be found here). By dividing the number of players in pro-ladder by the number of inhabitants (in millions) we can get the number of pro players per capita.

season country total_matches num_players pro_players_per_million matches_per_player
M2_08 Draconid 2020 Poland 72225 267 7.047129 270.505618
M2_08 Draconid 2020 Estonia 1726 7 5.280436 246.571429
M2_08 Draconid 2020 Russian Federation 195905 673 4.613626 291.092125
M2_08 Draconid 2020 Belarus 10260 39 4.125931 263.076923
M2_08 Draconid 2020 Ukraine 52333 162 3.682351 323.043210

The top 5 countries is comprised out of Eastern European Countries, which is no surprise as the company that created Gwent is based in Poland and The Witcher lore has been created based on Slavic myths and legends. Iceland, Finland, Hong Kong, Malta and Croatia complete the top 10. These are all relatively small countries, so a single player making it up to Pro Rank boosts them up in the ranking.

You can download the full table, which includes data for all seasons and countries here.

Which Country has the best Gwent Team

Now we know where the most pro players are per capita, but what if countries were able to send a team of three e-athletes to a world championship? Which countries would do best with their team of three pro players. To this end all countries with three or more players were selected, and the top 3 players for each of those countries picked. Next, the average MMR and total MMR for those players, was calculated as well as the efficiency to climb and the rank for each country. The code is up on GitHub for those interested, but also here it is a simple matter of filtering and grouping data using built-in pandas functions.

The results for Season of the Draconid are shown below. It seems that China has the best team of three this season followed by Russia and Poland.

season country mean_mmr total_mmr mean_matches_per_player total_matches nation_rank efficiency lei
M2_08 Draconid 2020 China 10489.333333 31468 409 1227 1 2.174409 43.974703
M2_08 Draconid 2020 Russian Federation 10479.666667 31439 636 1908 2 1.383124 34.881052
M2_08 Draconid 2020 Poland 10439.333333 31318 657 1971 3 1.277524 32.745512

Player Summaries

For players that made it up to Pro Rank during multiple seasons we’ll quickly generate a summary. Again the groupby and agg function are being leveraged again to group things and get the summary statistics. We’ll count the number of appearnaces on pro ladder, the min, mean, max MMR score. Average number of matches and total number of matches as well as the best global and national ranks.

This can give you a quick impression of all the data available on a player. Here you can see the output from myself (handle sepro).

name country appearances min_mmr mean_mmr max_mmr mean_matches num_matches best_rank best_national_rank
sepro Belgium 3 9746 9782 9820 243 728 1138 2.0

You can download the full table here to find your own or your favorite players stats.

Conclusion

Initially, I set out to get players national ranks. When you are from a small country just making it to Pro Rank will likely give you bragging rights about being in the top 3 of your country. Though with some fairly basic data science you can very quickly get a lot more details on various aspects of the game.

This type of project I would really recommend for people starting out with programming. Find a topic you like and write some code to get some data about it, do some analysis and generate a few plots.