Every so often while doing data analysis, I have come across a situation where I have two datasets, which have the same structure but with small difference between the two. For example:
In the above cases, I want to quickly know what has changed across the two data.frames, how much has changed, and also to get a quick summary of the extent of change. There are packages like the famous compare package on R, but that was focused more on the structure of the data frame and lesser on the data itself. I was not able to easily identify and isolate what has changed in the dataset. So I decided to write one for myself. That is what compareDF package is all about.
The package has a single function, compare_df. It takes in two data frames, and one or more grouping variables and does a comparison between the the two. In addition you can sepcificy columns to ignore, decide how many rows of changes to be displayed in the case of the HTML output, and decide what tolerance you want to provide to detect change.
Let’s take the example of a teacher who wants to compare the marks and grades of students across two years, 2010 and 2011. The data is stored in tabular format.
load("../../compareDF/data/results_2010.rda")
load("../../compareDF/data/results_2011.rda")
print(results_2010)## Division Student Maths Physics Chem Discipline PE Art
## 1 A Isaac 90 84 91 B B 34
## 2 A Akshay 85 92 91 B B 36
## 3 A Vishwas 93 93 92 A B 21
## 4 A Rohit 95 92 71 C B 37
## 5 A Venu 99 92 82 A E 78
## 6 A Ananth 99 81 91 B A 24
## 7 B Jojy 67 92 81 B A 27
## 8 B Bulla 84 73 81 C A 68
## 9 B Katti 90 95 99 C B 49
## 10 B Dhakkan 78 96 71 C C 39
## 11 B Macho 90 82 81 A+ D 30
## 12 B Mugger 95 71 94 A C 26print(results_2011)## Division Student Maths Physics Chem Discipline PE Art
## 1 A Isaac 90 84 91 A B 34
## 2 A Akshay 85 92 91 A B 36
## 3 A Vishwas 82 93 92 B B 21
## 4 A Rohit 94 92 71 D B 37
## 5 A Venu 100 92 82 A E 78
## 6 A Ananth 78 81 91 B A 24
## 7 B Jojy 99 92 81 B A 27
## 8 B Bulla 97 73 81 C A 68
## 9 B Katti 78 95 99 C B 49
## 10 B Rohit 79 96 71 C C 39
## 11 B Macho 90 82 81 A+ D 30
## 12 B Vikram 99 79 98 A B 99
## 13 B DIkChik 91 71 84 E C 99The data shows the performance of students in two divisions, A and B for two years. Some subjects like Maths, Physics, Chemistry and Art are given scores while others like Discipline and PE are given grades.
It is possible that there are students of the same name in two divisions, for example, there is a Rohit in both the divisions in 2011.
It is also possible that some students have dropped out, or added new across the two years. Eg: - Mugger and Dhakkan dropped out while Vikram and Dikchik where added in the Division B
The package allows a user to quickly identify these changes.
Now let’s compare the performance of the students across the years. The grouping variables is the Student columns. We will ignore the Division columns and assume that the student names are unique across divisions. In this sub-example, if a student appears in two division, he has studied in both of them.
library(compareDF)
ctable_student = compare_df(results_2011, results_2010, c("Student"))## Creating comparison table...## Loading required namespace: htmlTable## Creating HTML table for first 100 rowsctable_student$comparison_df## Student chng_type Division Maths Physics Chem Discipline PE Art
## 1 Akshay + A 85 92 91 A B 36
## 2 Akshay - A 85 92 91 B B 36
## 3 Ananth + A 78 81 91 B A 24
## 4 Ananth - A 99 81 91 B A 24
## 5 Bulla + B 97 73 81 C A 68
## 6 Bulla - B 84 73 81 C A 68
## 7 Dhakkan - B 78 96 71 C C 39
## 8 Isaac + A 90 84 91 A B 34
## 9 Isaac - A 90 84 91 B B 34
## 10 Jojy + B 99 92 81 B A 27
## 11 Jojy - B 67 92 81 B A 27
## 12 Katti + B 78 95 99 C B 49
## 13 Katti - B 90 95 99 C B 49
## 14 Mugger - B 95 71 94 A C 26
## 15 Rohit + A 94 92 71 D B 37
## 16 Rohit + B 79 96 71 C C 39
## 17 Rohit - A 95 92 71 C B 37
## 18 Venu + A 100 92 82 A E 78
## 19 Venu - A 99 92 82 A E 78
## 20 Vishwas + A 82 93 92 B B 21
## 21 Vishwas - A 93 93 92 A B 21
## 22 DIkChik + B 91 71 84 E C 99
## 23 Vikram + B 99 79 98 A B 99By default, no columns are excluded from the comparison, so any of the tuple of grouping variables which are different across the two data frames are shown in the comparison table. The comparison_df table shows all the rows for which at least one record has changed. If nothing has changed across the two tables, the rows are not displayed. If a new record has been introduced or a record has been removed, those are displayed as well.
For example, Akshay, Division A has the exact same scores but has two different grades for Discipline across the two years so that row is included.
However, Macho, Division B has had the exact same scores in both the years, so his data is not shown in the comparison table.
While the comparison, table can be quickly summarized in various forms for futher analysis, it is very difficult to process visually. The html_output provides a way to represent this is a way that is easier for the numan eye to read. NOTE: You need to install the htmlTable package for the HTML comparison to work. For the purpose of the readme I am attaching the html as a png because github markdown doesn’t retain styles.
print(ctable_student$html_output)| Student | chng_type | Division | Maths | Physics | Chem | Discipline | PE | Art |
|---|---|---|---|---|---|---|---|---|
| Akshay | + | A | 85 | 92 | 91 | A | B | 36 |
| Akshay | - | A | 85 | 92 | 91 | B | B | 36 |
| Ananth | + | A | 78 | 81 | 91 | B | A | 24 |
| Ananth | - | A | 99 | 81 | 91 | B | A | 24 |
| Bulla | + | B | 97 | 73 | 81 | C | A | 68 |
| Bulla | - | B | 84 | 73 | 81 | C | A | 68 |
| Dhakkan | - | B | 78 | 96 | 71 | C | C | 39 |
| Isaac | + | A | 90 | 84 | 91 | A | B | 34 |
| Isaac | - | A | 90 | 84 | 91 | B | B | 34 |
| Jojy | + | B | 99 | 92 | 81 | B | A | 27 |
| Jojy | - | B | 67 | 92 | 81 | B | A | 27 |
| Katti | + | B | 78 | 95 | 99 | C | B | 49 |
| Katti | - | B | 90 | 95 | 99 | C | B | 49 |
| Mugger | - | B | 95 | 71 | 94 | A | C | 26 |
| Rohit | + | A | 94 | 92 | 71 | D | B | 37 |
| Rohit | + | B | 79 | 96 | 71 | C | C | 39 |
| Rohit | - | A | 95 | 92 | 71 | C | B | 37 |
| Venu | + | A | 100 | 92 | 82 | A | E | 78 |
| Venu | - | A | 99 | 92 | 82 | A | E | 78 |
| Vishwas | + | A | 82 | 93 | 92 | B | B | 21 |
| Vishwas | - | A | 93 | 93 | 92 | A | B | 21 |
| DIkChik | + | B | 91 | 71 | 84 | E | C | 99 |
| Vikram | + | B | 99 | 79 | 98 | A | B | 99 |
Now it is very easy to see recognize what has changed. A single cell is colored if it has changed across the two datasets. The value of the cell in the older dataset is colored red and the value of the cell in the newer dataset is colored green. Cells that haven’t changed across the two datasets are colored grey.
If a new row was introduced, the Row group names (and all the other columns for that row as well ) are colored in Green. Similarly, a row group name (and the other columns in that row) are colored red if a row was removed.
For Example, Akshay, Ananth and Bulla has had changes in scored, which are in Discipline, Maths, and Maths respectively. Dhakkan and Mugger have dropped out of the dataset from 2010 and the full rows are shown in red, which DikChik and Vikram have joined new in the data set and the full rows are in green.
The same data is represented in tabular form (for further analysis, if necessary) in the comparison_table_diff object
ctable_student$comparison_table_diff## Student chng_type Division Maths Physics Chem Discipline PE Art
## 1 . + . . . . + . .
## 2 . - . . . . - . .
## 3 . + . + . . . . .
## 4 . - . - . . . . .
## 5 . + . + . . . . .
## 6 . - . - . . . . .
## 7 - - - - - - - - -
## 8 . + . . . . + . .
## 9 . - . . . . - . .
## 10 . + . + . . . . .
## 11 . - . - . . . . .
## 12 . + . + . . . . .
## 13 . - . - . . . . .
## 14 - - - - - - - - -
## 15 . + + + + . + + +
## 16 . + + + + . + + +
## 17 . - - - - . - - -
## 18 . + . + . . . . .
## 19 . - . - . . . . .
## 20 . + . + . . + . .
## 21 . - . - . . - . .
## 22 + + + + + + + + +
## 23 + + + + + + + + +You can get an details of what has changed for each group using the change_count object in the output. A summary of the same is provided in the change_summary object.
ctable_student$change_count## Source: local data frame [13 x 4]
##
## Student changes additions removals
## (fctr) (dbl) (dbl) (dbl)
## 1 Akshay 1 0 0
## 2 Ananth 1 0 0
## 3 Bulla 1 0 0
## 4 Dhakkan 0 1 0
## 5 Isaac 1 0 0
## 6 Jojy 1 0 0
## 7 Katti 1 0 0
## 8 Mugger 0 1 0
## 9 Rohit 1 0 1
## 10 Venu 1 0 0
## 11 Vishwas 1 0 0
## 12 DIkChik 0 0 1
## 13 Vikram 0 0 1ctable_student$change_summary## old_obs new_obs changes additions removals
## 12 13 9 2 3We can also group_multiple columns into the grouping variable
ctable_student_div = compare_df(results_2011, results_2010, c("Division", "Student"))## Grouping grouping columns## Creating comparison table...## Creating HTML table for first 100 rowsctable_student_div$comparison_dfgrp chng_type Division Student Maths Physics Chem Discipline PE Art 1 1 + A Akshay 85 92 91 A B 36 2 1 - A Akshay 85 92 91 B B 36 3 2 + A Ananth 78 81 91 B A 24 4 2 - A Ananth 99 81 91 B A 24 5 3 + A Isaac 90 84 91 A B 34 6 3 - A Isaac 90 84 91 B B 34 7 4 + A Rohit 94 92 71 D B 37 8 4 - A Rohit 95 92 71 C B 37 9 5 + A Venu 100 92 82 A E 78 10 5 - A Venu 99 92 82 A E 78 11 6 + A Vishwas 82 93 92 B B 21 12 6 - A Vishwas 93 93 92 A B 21 13 7 + B Bulla 97 73 81 C A 68 14 7 - B Bulla 84 73 81 C A 68 15 8 + B DIkChik 91 71 84 E C 99 16 9 + B Jojy 99 92 81 B A 27 17 9 - B Jojy 67 92 81 B A 27 18 10 + B Katti 78 95 99 C B 49 19 10 - B Katti 90 95 99 C B 49 20 12 + B Rohit 79 96 71 C C 39 21 13 + B Vikram 99 79 98 A B 99 22 14 - B Dhakkan 78 96 71 C C 39 23 15 - B Mugger 95 71 94 A C 26
Now Rohits in each individual division are considered as belonging to separate groups and compared accordingly. All the other summaries also change appropriately.
You can ignore certain columns using the exclude parameter. The fields that have to be excluded can be given as a character vector. (This is a convenience function to deal with the case where some columns are not included)
For dataframes which have a large amount of differences in them, generating HTML might take a long time and crash your system. So the maximum diff size for the HTML (and for the HTML visualization only) is capped at 100 by default. If you want to see more difference, you can change the limit_html parameter appropriately.
It is possible that you’d like numbers very close to each other to be ignored. For example, if the marks of a student vary by less that 5% across the years, it could be due to random error and not any real performance indictaor. In such a case, you would want to give a tolerance parameter. For this case, giving a tolerance of 0.05 would ignore all the changes that are less than 5% apart from the lower value.
ctable_student_div = compare_df(results_2011, results_2010, c("Division", "Student"), tolerance = 0.05)## Grouping grouping columns## Creating comparison table...## Creating HTML table for first 100 rowsctable_student_div$html_output| grp | chng_type | Division | Student | Maths | Physics | Chem | Discipline | PE | Art |
|---|---|---|---|---|---|---|---|---|---|
| 1 | + | A | Akshay | 85 | 92 | 91 | A | B | 36 |
| 1 | - | A | Akshay | 85 | 92 | 91 | B | B | 36 |
| 2 | + | A | Ananth | 78 | 81 | 91 | B | A | 24 |
| 2 | - | A | Ananth | 99 | 81 | 91 | B | A | 24 |
| 3 | + | A | Isaac | 90 | 84 | 91 | A | B | 34 |
| 3 | - | A | Isaac | 90 | 84 | 91 | B | B | 34 |
| 4 | + | A | Rohit | 94 | 92 | 71 | D | B | 37 |
| 4 | - | A | Rohit | 95 | 92 | 71 | C | B | 37 |
| 6 | + | A | Vishwas | 82 | 93 | 92 | B | B | 21 |
| 6 | - | A | Vishwas | 93 | 93 | 92 | A | B | 21 |
| 7 | + | B | Bulla | 97 | 73 | 81 | C | A | 68 |
| 7 | - | B | Bulla | 84 | 73 | 81 | C | A | 68 |
| 8 | + | B | DIkChik | 91 | 71 | 84 | E | C | 99 |
| 9 | + | B | Jojy | 99 | 92 | 81 | B | A | 27 |
| 9 | - | B | Jojy | 67 | 92 | 81 | B | A | 27 |
| 10 | + | B | Katti | 78 | 95 | 99 | C | B | 49 |
| 10 | - | B | Katti | 90 | 95 | 99 | C | B | 49 |
| 12 | + | B | Rohit | 79 | 96 | 71 | C | C | 39 |
| 13 | + | B | Vikram | 99 | 79 | 98 | A | B | 99 |
| 14 | - | B | Dhakkan | 78 | 96 | 71 | C | C | 39 |
| 15 | - | B | Mugger | 95 | 71 | 94 | A | C | 26 |
Venu from division A who had a score change from 100 to 99 is no longer present in the diff calculation or in the output
Naturally, tolerance has no meaning for non-numeric values.
I’d like to thank System Insights Inc. for all the things that I have learned while working there which I have used one way or the other in this package.
A version of this blog has been published as the README for the package.