NB The main notebook of model creation is found in this directory as model_creation.ipynb.
BoardGameGeek.com (BGG) is a boardgame database website that includes tens of thousands of entries. The data for each game includes the game's complexity, rating, designer, publisher, among many other points. Games are ranked according to BGG own ranking system. Further, they provide a 'Geek' rating alongside their user provided rating average to evaluate ranking.
Objective: Provide insight to boardgame designers and publishers regarding the prospective ranking of their game through a supervised learning model.
I created a scraping solution using BeautifulSoup and Selenium to scrape 10,000 entries in BGG.
The use of Selenium was necessary to collect market price data from dynamic marketplace entries.
The scraped data can be found in this repo as a .csv file for further analysis.
In toto:
Int64Index: 10018 entries, 0 to 10018
Data columns (total 113 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 link 10018 non-null object
1 rank 10018 non-null float64
2 desc 10018 non-null object
3 geek 10018 non-null float64
4 rating 10018 non-null float64
5 num_voters 10018 non-null int64
6 price 10018 non-null object
7 average 10018 non-null object
8 avgweight 10018 non-null float64
9 baverage 10018 non-null float64
10 maxplayers 10018 non-null float64
11 maxplaytime 10018 non-null float64
12 minage 10018 non-null float64
13 minplayers 10018 non-null float64
14 minplaytime 10018 non-null float64
15 numcomments 10018 non-null float64
16 numfans 10018 non-null float64
17 numgeeklists 10018 non-null float64
18 numhasparts 10018 non-null float64
19 numowned 10018 non-null float64
20 numplays 10018 non-null float64
21 numplays_month 10018 non-null float64
22 numprevowned 10018 non-null float64
23 numtrading 10018 non-null float64
24 numwanting 10018 non-null float64
25 numwantparts 10018 non-null float64
26 numweights 10018 non-null float64
27 numwish 10018 non-null float64
28 numwishlistcomments 10018 non-null float64
29 playmonth 10018 non-null object
30 price_mkt 3332 non-null float64
31 stddev 10018 non-null float64
32 usersrated 10018 non-null float64
33 views 10018 non-null float64
34 walmart_price 260 non-null object
35 yearpublished 10018 non-null float64
36 genres 10009 non-null object
37 designer 9906 non-null object
38 publish 10008 non-null object
39 strategy_war 10018 non-null int64
40 strategy 10018 non-null int64
41 family 10018 non-null int64
42 thematic 10018 non-null int64
43 war 10018 non-null int64
44 party 10018 non-null int64
45 abstract 10018 non-null int64
46 rank_des 9906 non-null float64
47 geek_des 9906 non-null float64
48 rating_des 9906 non-null float64
49 num_voters_des 9906 non-null float64
50 baverage_des 9906 non-null float64
51 maxplayers_des 9906 non-null float64
52 maxplaytime_des 9906 non-null float64
53 minage_des 9906 non-null float64
54 minplayers_des 9906 non-null float64
55 minplaytime_des 9906 non-null float64
56 numcomments_des 9906 non-null float64
57 numfans_des 9906 non-null float64
58 numgeeklists_des 9906 non-null float64
59 numowned_des 9906 non-null float64
60 numplays_des 9906 non-null float64
61 numplays_month_des 9906 non-null float64
62 numtrading_des 9906 non-null float64
63 numwanting_des 9906 non-null float64
64 numwantparts_des 9906 non-null float64
65 numweights_des 9906 non-null float64
66 numwish_des 9906 non-null float64
67 price_mkt_des 7238 non-null float64
68 views_des 9906 non-null float64
69 strategy_war_des 9906 non-null float64
70 strategy_des 9906 non-null float64
71 family_des 9906 non-null float64
72 thematic_des 9906 non-null float64
73 war_des 9906 non-null float64
74 party_des 9906 non-null float64
75 abstract_des 9906 non-null float64
76 numdes 9906 non-null float64
77 rank_pub 10008 non-null float64
78 geek_pub 10008 non-null float64
79 rating_pub 10008 non-null float64
80 num_voters_pub 10008 non-null float64
81 baverage_pub 10008 non-null float64
82 maxplayers_pub 10008 non-null float64
83 maxplaytime_pub 10008 non-null float64
84 minage_pub 10008 non-null float64
85 minplayers_pub 10008 non-null float64
86 minplaytime_pub 10008 non-null float64
87 numcomments_pub 10008 non-null float64
88 numfans_pub 10008 non-null float64
89 numgeeklists_pub 10008 non-null float64
90 numowned_pub 10008 non-null float64
91 numplays_pub 10008 non-null float64
92 numplays_month_pub 10008 non-null float64
93 numtrading_pub 10008 non-null float64
94 numwanting_pub 10008 non-null float64
95 numwantparts_pub 10008 non-null float64
96 numweights_pub 10008 non-null float64
97 numwish_pub 10008 non-null float64
98 price_mkt_pub 8604 non-null float64
99 views_pub 10008 non-null float64
100 strategy_war_pub 10008 non-null float64
101 strategy_pub 10008 non-null float64
102 family_pub 10008 non-null float64
103 thematic_pub 10008 non-null float64
104 war_pub 10008 non-null float64
105 party_pub 10008 non-null float64
106 abstract_pub 10008 non-null float64
107 numpub 10008 non-null float64
108 yearpublished_dt 9964 non-null datetime64[ns]
109 diff_des 9906 non-null float64
110 diff_pub 10008 non-null float64
111 quality_des 10018 non-null int64
112 quality_pub 10018 non-null int64 Columns with suffix des represent mean values for a designer appended to each game they designed. Likewise, pub represents mean publisher data.
Genre columns are encoded with dummy variables (0,1).
quality entries include dummy variables with 1 being average or above average for designer/publisher.
The following heatmap demonstrates the features I chose to focus on.
The model focuses on the following features :
sub_df[['avgweight', 'strategy','war','family', 'abstract', 'thematic', 'geek_des', 'geek_pub']]
sub_df['log_geek_des'] = np.log(sub_df['geek_des'])
sub_df['log_geek_pub'] = np.log(sub_df['geek_pub'])Our target/response variable is the ranking on BGG.
The R2 of this model is 0.60 .
Mean Absolute Error:
'Train MAE:', 0.4198582356908472, 'Test MAE:', 0.40880771810103766
The above model scores 0.61 on test data. I was able to eliminate negative output variables by taking the log of the output and using a polynomial fit.
Better pricing information would be helpful. As many games sold on marketplaces are collector items, pricing contains numerous outliers.