Skip to content

Instantly share code, notes, and snippets.

Last active April 24, 2024 18:22
Show Gist options
  • Save conormm/fd8b1980c28dd21cfaf6975c86c74d07 to your computer and use it in GitHub Desktop.
Save conormm/fd8b1980c28dd21cfaf6975c86c74d07 to your computer and use it in GitHub Desktop.
R to Python: Data wrangling with dplyr and pandas

R to python data wrangling snippets

The dplyr package in R makes data wrangling significantly easier. The beauty of dplyr is that, by design, the options available are limited. Specifically, a set of key verbs form the core of the package. Using these verbs you can solve a wide range of data problems effectively in a shorter timeframe. Whilse transitioning to Python I have greatly missed the ease with which I can think through and solve problems using dplyr in R. The purpose of this document is to demonstrate how to execute the key dplyr verbs when manipulating data using Python (with the pandas package).

dplyr is organised around six key verbs:

  • filter : subset a dataframe according to condition(s) in a variable(s)
  • select : choose a specific variable or set of variables
  • arrange : order dataframe by index or variable
  • group_by : create a grouped dataframe
  • summarise : reduce variable to summary variable (e.g. mean)
  • mutate : transform dataframe by adding new variables

The excellent pandas package in Python easily allows you to implement all of these actions (and much, much more!). Below are some snippets to highlight some of the more basic conversions.


June 8th 2018 update: All of this code still works in pandas and should ease the transition from R, but for those interested in getting the most out of the package I strongly recommend this series on modern pandas



filter(df, var > 20000 & var < 30000) 
filter(df, var == 'string') # df %>% filter(var != 'string')
df %>% filter(var != 'string')
df %>% group_by(group) %>% filter(sum(var) > 2000000)


df[(df['var'] > 20000) & (df['var'] < 30000)]
df[df['var'] == 'string']
df[df['var'] != 'string']
df.groupby('group').filter(lambda x: sum(x['var']) > 2000000)



select(df, var1, var2)
select(df, -var3)


df[['var1', 'var2']]
df.drop('var3', 1)



arrange(df, var1)
arrange(df, desc(var1))


df.sort_values('var1', ascending=False)



df %>% group_by(group) 
df %>% group_by(group1, group2)
df %>% ungroup()


df.groupby(['group1', 'group2'])
df.reset_index() / or when grouping: df.groupby('group1', as_index=False)

Summarise / Aggregate df by group


df %>% group_by(group) %>% summarise(mean_var1 = mean(var1))
df %>% 
  group_by(group1, group2) %>% 
  summarise(mean_var1 = mean(var1), 
            sum_var1 = sum(var1), 
            count_var1 = n())
df %>% 
  group_by(group1, group2) %>% 
  summarise(mean_var1 = mean(var1),
            sum_2 = sum(var2),
            var3 = first(var3))


df.groupby('group1')['var1'].agg({'mean_col' : np.mean()}) # pass dict to specifiy column name

df.groupby(['group1', 'group2'])['var1].agg(['mean', 'sum', 'count']) # for count also consider 'size'. size will return n for NaN values also, whereas 'count' will not.

# first perform the aggregation
group_agg = df.groupby(["group1", "group2"]).agg({
  "var1" : ["mean"], 
  "var2" : ["sum"], 
  "var3" : ["first"]
# second rename the columns by joining the column name with the agg function (e.g. "var1_mean")
group_agg.columns = ["_".join(x) for x in group_agg.columns.ravel()]

# You can also pass multiple functions to aggregate the same column e.g:
group_agg = df.groupby(["group1", "group2"]).agg({"var1" : ["mean", "std", "sum"]})

Mutate / transform df by group


df %>% group_by(group) %>% mutate(mean_var1 = mean(var1))


df.groupby('group').assign(mean_var1 = lambda x: np.mean(x.var1)



df %>% distinct()
df %>% distinct(col1) # returns dataframe with unique values of col1


df.drop_duplicates(subset='col1') # returns dataframe with unique values of col1



sample_n(df, 100)
sample_frac(df, 0.5)


Copy link

Thank you so much! @epetrovski do you think you could give an example of how to group and mutate using .groupby() and .apply()? I'm also running into the issue where .groupby() and .assign() don't work together.
I ended up doing something along these lines:
df['mean_var1']=df.groupby('group').apply(lambda x: np.mean(x.var1)).reset_index(level=0, drop=True)
But am not sure if there is an easier way, and if this is what you had in mind. Thanks!

That would be the way to do it with .apply(). The problem is, though, that it's a bit unsafe since using .reset_index() means that you're assigning back to the data frame without keeping track of the index. This will be a problem if you return fewer rows than you input for some reason.
A safer and faster way that I've found is to use the .pipe() method like this:

df['mean_var1'] = df.groupby('group').pipe(lambda x: x.var1.transform('mean'))

You dont need pipe for this, simply transform the column and assign to the new column name :

   df['mean_var1'] = df.groupby('group').var1.transform('mean')

Fair point. But using .pipe() will allow you to transform several columns at once making it analogous to mutate() which the post is about.

I respectfully disagree. I'll show some examples below on what I believe should be the way in Pandas to do it; please correct me where the code is wrong or suboptimal :

print(df )
      tag     val1    val2
0	B	0	0
1	A	1	2
2	A	2	4
3	B	3	6
4	B	4	8

grp = df.groupby("tag")

Let's create new columns and assign to the dataframe :

	tag	val1	val2	val1_mean	val2_sum
0	B	0	0	2.333333	14
1	A	1	2	1.500000	6
2	A	2	4	1.500000	6
3	B	3	6	2.333333	14
4	B	4	8	2.333333	14

If you are applying the same function to the columns, you can use the unpacking method :
df[["val1_mean", "val2_mean"]] = grp[["val1", "val2"]].transform("mean")


tag	val1	val2	val1_mean	val2_mean
0	B	0	0	2.333333	4.666667
1	A	1	2	1.500000	3.000000
2	A	2	4	1.500000	3.000000
3	B	3	6	2.333333	4.666667
4	B	4	8	2.333333	4.666667

I do not see the need for pipe for these scenarios. Pipe, in my opinion, comes into play when you need to reuse the groupby object. like
grp.pipe(lambda x: x.sum() - x.mean()).

The groupby object is reused in the sum and mean before the difference is applied. I dont see the effectiveness of calling transform inside pipe when transform can do the job easily. Of course, you may have a use case where pipe does the job better than transform - I would love to learn about that. Cheers.

Copy link

@samukweku, you say that "Pipe, in my opinion, comes into play when you need to reuse the groupby object."

So in my use case I need to do something like this, where I take the sum of one variable and add to it the minimum of another but using the same grouping:

df['res'] = df.groupby('id').pipe(lambda x: x.var1.transform('sum') + x.var2.transform('min'))

Anyways, this is going into too much depth here. I think it's fair to say that there are several ways of accomplishing a group_by() %>% mutate() in pandas but df.groupby('group').assign(mean_var1 = lambda x: np.mean(x.var1) which the post suggests isn't one of them and that's the most important point.

Copy link

Ahh! @epetrovski, totally correct there. Your example makes sense. Cheers.

Copy link

df.groupby('group').assign(mean_var1 = lambda x: np.mean(x.var1)
would result in
AttributeError: 'DataFrameGroupBy' object has no attribute 'assign'
Can you update your code? @conormm

Copy link

pwwang commented Apr 30, 2021

datar ports dplyr and other packages to python for you.

It follows the original R packages' API design.

With datar, you can do:

from datar import f
from datar.dplyr import group_by, mutate
from datar.base import mean

df >> group_by( >> mutate(mean_var1 = mean(f.var1))
# compared to the R syntax:
# df %>% group_by(group) %>% mutate(mean_var1 = mean(var1))

Copy link

roboes commented Jun 27, 2021

For "Mutate / transform df by group" I do the following:

# df example
df = pd.DataFrame(data = {'group': ['A', 'A', 'B', 'B'], 'var1': [1, 4, 2, 2]})

# Mutate / transform df by group
df.assign(mean_var1 = lambda x: x.groupby('group')['var1'].transform('mean'))

Copy link

@RobertoG09, i think selecting var1 before transform might be faster, since you are transforming one column, as against all columns before selecting your traget column : df.assign(mean_var1 = lambda x: x.groupby('group')['var1'].transform('mean'))

Copy link

roboes commented Aug 3, 2021

@samukweku this makes sense. I have updated the original comment to reflect your feedback. Thanks!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment