Sarepta Self Assessment

Priyanka Gagneja
01/17/2020

Beer Data Analysis

  • Introduction to Data
  • Data Quality Review
  • Data Exploration
  • Answers to the Questions asked

Beer Data Analysis - Introduction to Data

  • 1.5 mn beer reviews

  • Reviews are measured on 11-point (Likert) scale varying between 0 - 5 with focus on aspects such as:

    • appearance,
    • aroma,
    • palate,
    • taste, and
    • overall impression

  • 5840 Breweries

  • 66K Beers

  • 104 Beer Styles

Beer Data Analysis - Data Quality Review

library(DataExplorer)
introduce(beer)

     rows columns discrete_columns continuous_columns all_missing_columns
1 1586614      13                4                  9                   0
  total_missing_values complete_rows total_observations memory_usage
1                67785       1518829           20625982    128171432

Note : Missingness of data does not seem to be any issue with this data. All the fields are complete except for beer_abv field that has 67K NAs (4.2%).

library(inspectdf)
inspect_na(beer)

# A tibble: 13 x 3
   col_name             cnt  pcnt
   <chr>              <int> <dbl>
 1 beer_abv           67785  4.27
 2 brewery_id             0  0   
 3 brewery_name           0  0   
 4 review_time            0  0   
 5 review_overall         0  0   
 6 review_aroma           0  0   
 7 review_appearance      0  0   
 8 review_profilename     0  0   
 9 beer_style             0  0   
10 review_palate          0  0   
11 review_taste           0  0   
12 beer_name              0  0   
13 beer_beerid            0  0

Beer Data Analysis - Data Exploration

inspect_types(beer)

# A tibble: 3 x 4
  type      cnt  pcnt col_name 
  <chr>   <int> <dbl> <list>   
1 numeric     6  46.2 <chr [6]>
2 factor      4  30.8 <chr [4]>
3 integer     3  23.1 <chr [3]>

Beer Data Analysis - Data Exploration

inspect_num(beer)

# A tibble: 9 x 10
  col_name     min     q1 median   mean     q3    max      sd pcnt_na hist 
  <chr>      <dbl>  <dbl>  <dbl>  <dbl>  <dbl>  <dbl>   <dbl>   <dbl> <nam>
1 brewery… 1.00e+0 1.43e2 4.29e2 3.13e3 2.37e3 2.80e4 5.58e+3    0    <tib…
2 review_… 8.41e+8 1.17e9 1.24e9 1.22e9 1.29e9 1.33e9 7.65e+7    0    <tib…
3 review_… 0.      3.50e0 4.00e0 3.82e0 4.50e0 5.00e0 7.21e-1    0    <tib…
4 review_… 1.00e+0 3.50e0 4.00e0 3.74e0 4.00e0 5.00e0 6.98e-1    0    <tib…
5 review_… 0.      3.50e0 4.00e0 3.84e0 4.00e0 5.00e0 6.16e-1    0    <tib…
6 review_… 1.00e+0 3.50e0 4.00e0 3.74e0 4.00e0 5.00e0 6.82e-1    0    <tib…
7 review_… 1.00e+0 3.50e0 4.00e0 3.79e0 4.50e0 5.00e0 7.32e-1    0    <tib…
8 beer_abv 1.00e-2 5.20e0 6.50e0 7.04e0 8.50e0 5.77e1 2.32e+0    4.27 <tib…
9 beer_be… 3.00e+0 1.72e3 1.39e4 2.17e4 3.94e4 7.73e4 2.18e+4    0    <tib…

Beer Data Analysis - Data Exploration

show_plot(inspect_num(beer))

plot of chunk unnamed-chunk-6

Beer Data Analysis - Data Exploration

#show_plot(inspect_cat(beer)) 
inspect_cat(beer)

# A tibble: 4 x 5
  col_name          cnt common                  common_pcnt levels         
  <chr>           <int> <chr>                         <dbl> <named list>   
1 beer_name       56857 90 Minute IPA                 0.207 <tibble [56,85…
2 beer_style        104 American IPA                  7.41  <tibble [104 ×…
3 brewery_name     5743 Boston Beer Company (S…       2.49  <tibble [5,743…
4 review_profile… 33388 northyorksammy                0.367 <tibble [33,38…

Note: _name & _id variables need some cleaning up.

Beer Data Analysis - Quick Insights

# basic stats about each field
summary(beer)

# some more information with describe()
library(Hmisc)
describe(beer)
  • Most reviewed Breweries
    • Boston Beer Company (Samuel Adams),
    • Dogfish Head Brewery and
    • Stone Brewing Co.;
  • Most reviewed Beers
    • 90 Minute IPA,
    • India Pale Ale,
    • Old Rasputin Russian Imperial Stout;
  • Most reviewed Beer Styles
    • American IPA,
    • American Double / Imperial IPA,
    • American Pale Ale (APA);

Finally Answers to the Questions

Q-1. Which brewery produces the strongest beers by ABV% ?

Context:

  • How do we define 'strongest'
  • Alcohol by Volume %

plot of chunk unnamed-chunk-9

Challenge:

  • Different Breweries brew different no. of Beers.
  • No. of Reviews varies for different Beers. plot of chunk unnamed-chunk-10

Q-1. Which brewery produces the strongest beers by ABV% ?

How beer_abv values vary?

plot of chunk unnamed-chunk-11

Individual values

plot of chunk unnamed-chunk-12

Q-1. Which brewery produces the strongest beers by ABV% ?

plot of chunk unnamed-chunk-13

Brewery with a Beer with highest actual beer_abv

beer %>% arrange(desc(beer_abv)) %>% 
         select(brewery_name, beer_abv) %>% 
         top_n(1)

  brewery_name beer_abv
1 Schorschbräu     57.7

plot of chunk unnamed-chunk-15

Brewery with a Beer with highest average beer_abv

beer %>% select(brewery_name, beer_name, beer_abv) %>% 
         group_by(brewery_name) %>% 
         summarise(avg_abv = mean(beer_abv)) %>% 
         arrange(desc(avg_abv)) %>% 
         top_n(1)

# A tibble: 1 x 2
  brewery_name avg_abv
  <fct>          <dbl>
1 Schorschbräu    19.2

Q-2. If you had to pick 3 beers to recommend using only this data, which would you pick?

  • Simple Approach - Best Seller (i.e. Most/Best Reviewed)
    • highest overall rating
beer %>% select(beer_name, review_overall) %>% 
         arrange(desc(review_overall)) %>% 
         unique() %>% 
         head()

                   beer_name review_overall
1        Caldera Ginger Beer              5
2              Rauch Ür Bock              5
5           Caldera Pale Ale              5
14         Pilot Rock Porter              5
15          Vas Deferens Ale              5
16 Old Growth Imperial Stout              5

Q-2. If you had to pick 3 beers to recommend using only this data, which would you pick?

  • Simple Approach (Amazon)- Best Seller (i.e. Most/Best Reviewed)
    • maximum votes of overall rating of 5 
beer %>% select(beer_name, review_overall) %>% 
         group_by(beer_name, review_overall) %>% 
         summarise(count = n()) %>% 
         arrange(desc(review_overall),desc(count)) %>% 
         top_n(1) %>% 
         head()

# A tibble: 6 x 3
# Groups:   beer_name [6]
  beer_name                 review_overall count
  <fct>                              <dbl> <int>
1 Pliny The Elder                        5  1067
2 Trappist Westvleteren 12               5   594
3 Pliny The Younger                      5   293
4 Portsmouth Kate The Great              5   187
5 Live Oak HefeWeizen                    5   137
6 Citra DIPA                             5   111

Q-2. If you had to pick 3 beers to recommend using only this data, which would you pick?

  • Advanced Approach
    • Recommendation System
      • (Spotify) Collaborative filtering using cosine similarity - item-item or user-item similarity
      • Association Rules
      • Simple Value Decomposition

Q-3. Which of the factors (aroma, taste, appearance, palette) are most important in determining the overall quality of a beer?

Lets find out the relationship/correlations various review related variables 

library(corrplot)
beer_mat <- as.matrix(beer[,c("review_overall",       "review_aroma","review_palate","review_appearance","review_taste")])

corrplot(cor(beer_mat), method="number", type = "upper")

plot of chunk unnamed-chunk-19

Q-4. Lastly, if I typically enjoy a beer due to its aroma and appearance, which beer style should I try?

  • Simple Approach - Best Seller (i.e. Most/Best Reviewed)
    • maximum matches on aroma & appearance rating and overall_rating.
beer %>% select(beer_name, review_aroma, review_appearance, review_overall ) %>% 
         group_by(beer_name, review_aroma, review_appearance, review_overall) %>% 
         summarise(count = n()) %>% 
         arrange(desc(review_aroma), desc(review_appearance), desc(review_overall), desc(count)) %>% 
         top_n(1) %>% 
         head()

# A tibble: 6 x 5
# Groups:   beer_name, review_aroma, review_appearance [6]
  beer_name              review_aroma review_appearan… review_overall count
  <fct>                         <dbl>            <dbl>          <dbl> <int>
1 Pliny The Elder                   5                5              5   214
2 Trappist Westvleteren…            5                5              5   159
3 Founders KBS (Kentuck…            5                5              5   148
4 The Abyss                         5                5              5   126
5 Pliny The Younger                 5                5              5   109
6 Founders Breakfast St…            5                5              5   105

Thank You !!

Questions