QI have a data frame of 4 columns. I need to create a {key: value} dictionary for 2 of those columns where this {key: value} pair should be created for each separate line in the data frame. Please refer to the example below:
df>>
a b c d
0 1 2 3 4
1 9 8 7 6
Expected output>>
a b c d new-column
0 1 2 3 4 {a:1, b:2}
1 9 8 7 6 {a:9, b:8}
You can use to_dict, craft a Series and join to the original DataFrame:
df2 = df.join(pd.Series(df[['a','b']].to_dict('index'), name='new_column'))
Output:
a b c d new_column
0 1 2 3 4 {'a': 1, 'b': 2}
1 9 8 7 6 {'a': 9, 'b': 8}
You can assign the values of returned dictionary of to_dict('index') to new column
df['new-column'] = df[['a','b']].to_dict('index').values()
print(df)
a b c d new-column
0 1 2 3 4 {'a': 1, 'b': 2}
1 9 8 7 6 {'a': 9, 'b': 8}
Related
Here is the Dataframe I am working with:
You can create it using the snippet:
my_dict = {'id': [1,2,1,2,1,2,1,2,1,2,3,1,3, 3],
'category':['a', 'a', 'b', 'b', 'b', 'b', 'a', 'a', 'b', 'b', 'b', 'a', 'a', 'a'],
'value' : [1, 12, 34, 12, 12 ,34, 12, 35, 34, 45, 65, 55, 34, 25]
}
x = pd.DataFrame(my_dict)
x
I want to filter IDs based on the condition: for category a, the count of values should be 2 and for category b, the count of values should be 3. Therefore, I would remove id 1 from category a and id 3 from category b from my original dataset x.
I can write the code for individual categories and start removing id's manually by using the code:
x.query('category == "a"').groupby('id').value.count().loc[lambda x: x != 2]
x.query('category == "b"').groupby('id').value.count().loc[lambda x: x != 3]
But, I don't want to do it manually since there are multiple categories. Is there a better way of doing it by considering all the categories at once and remove id's based on the condition listed in a list/dictionary?
If need filter MultiIndex Series - s by dictionary use Index.get_level_values with Series.map and get equal values per groups in boolean indexing:
s = x.groupby(['category','id']).value.count()
d = {'a': 2, 'b': 3}
print (s[s.eq(s.index.get_level_values(0).map(d))])
category id
a 2 2
3 2
b 1 3
2 3
Name: value, dtype: int64
If need filter original DataFrame:
s = x.groupby(['category','id'])['value'].transform('count')
print (s)
0 3
1 2
2 3
3 3
4 3
5 3
6 3
7 2
8 3
9 3
10 1
11 3
12 2
13 2
Name: value, dtype: int64
d = {'a': 2, 'b': 3}
print (x[s.eq(x['category'].map(d))])
id category value
1 2 a 12
2 1 b 34
3 2 b 12
4 1 b 12
5 2 b 34
7 2 a 35
8 1 b 34
9 2 b 45
12 3 a 34
13 3 a 25
This question already has answers here:
How can I replicate rows of a Pandas DataFrame?
(10 answers)
Closed 11 months ago.
I want to replicate rows in a Pandas Dataframe. Each row should be repeated n times, where n is a field of each row.
import pandas as pd
what_i_have = pd.DataFrame(data={
'id': ['A', 'B', 'C'],
'n' : [ 1, 2, 3],
'v' : [ 10, 13, 8]
})
what_i_want = pd.DataFrame(data={
'id': ['A', 'B', 'B', 'C', 'C', 'C'],
'v' : [ 10, 13, 13, 8, 8, 8]
})
Is this possible?
You can use Index.repeat to get repeated index values based on the column then select from the DataFrame:
df2 = df.loc[df.index.repeat(df.n)]
id n v
0 A 1 10
1 B 2 13
1 B 2 13
2 C 3 8
2 C 3 8
2 C 3 8
Or you could use np.repeat to get the repeated indices and then use that to index into the frame:
df2 = df.loc[np.repeat(df.index.values, df.n)]
id n v
0 A 1 10
1 B 2 13
1 B 2 13
2 C 3 8
2 C 3 8
2 C 3 8
After which there's only a bit of cleaning up to do:
df2 = df2.drop("n", axis=1).reset_index(drop=True)
id v
0 A 10
1 B 13
2 B 13
3 C 8
4 C 8
5 C 8
Note that if you might have duplicate indices to worry about, you could use .iloc instead:
df.iloc[np.repeat(np.arange(len(df)), df["n"])].drop("n", axis=1).reset_index(drop=True)
id v
0 A 10
1 B 13
2 B 13
3 C 8
4 C 8
5 C 8
which uses the positions, and not the index labels.
You could use set_index and repeat
In [1057]: df.set_index(['id'])['v'].repeat(df['n']).reset_index()
Out[1057]:
id v
0 A 10
1 B 13
2 B 13
3 C 8
4 C 8
5 C 8
Details
In [1058]: df
Out[1058]:
id n v
0 A 1 10
1 B 2 13
2 C 3 8
It's something like the uncount in tidyr:
https://tidyr.tidyverse.org/reference/uncount.html
I wrote a package (https://github.com/pwwang/datar) that implements this API:
from datar import f
from datar.tibble import tribble
from datar.tidyr import uncount
what_i_have = tribble(
f.id, f.n, f.v,
'A', 1, 10,
'B', 2, 13,
'C', 3, 8
)
what_i_have >> uncount(f.n)
Output:
id v
0 A 10
1 B 13
1 B 13
2 C 8
2 C 8
2 C 8
Not the best solution, but I want to share this: you could also use pandas.reindex() and .repeat():
df.reindex(df.index.repeat(df.n)).drop('n', axis=1)
Output:
id v
0 A 10
1 B 13
1 B 13
2 C 8
2 C 8
2 C 8
You can further append .reset_index(drop=True) to reset the .index.
Let's say I've got a pandas dataframe that looks like:
df1 = pd.DataFrame({"Item ID":["A", "B", "C", "D", "E"], "Value1":[1, 2, 3, 4, 0],
"Value2":[4, 5, 1, 8, 7], "Value3":[3, 8, 1, 2, 0],"Value4":[4, 5, 7, 9, 4]})
print(df1)
Item_ID Value1 Value2 Value3 Value4
0 A 1 4 3 4
1 B 2 5 8 5
2 C 3 1 1 7
3 D 4 8 2 9
4 E 0 7 0 4
Now I've got a second dataframe that looks like:
df2 = {"Item ID":["A", "C", "D"], "Value5":[4, 5, 7]}
print(df2)
Item_ID Value5
0 A 4
1 C 5
2 D 7
What I want do is find where the Item ID's match between my two data frames, and then add the "Value5" column values to the intersection of the rows AND ONLY columns Value1 and Value2 from df1 (these columns could change every iteration, so these columns need to be contained in a variable).
My output should show:
4 added to Row A, columns "Value1" and "Value2"
5 added to Row C, columns "Value1" and "Value2"
7 added to Row D, columns "Value1" and "Value2"
Item_ID Value1 Value2 Value3 Value4
0 A 5 8 3 4
1 B 2 5 8 5
2 C 8 6 1 7
3 D 11 15 2 9
4 E 0 7 0 4
Of course my data is many thousand rows long. I can do it using a for loop, but this is taking way too long. I want to be able to vectorize this in some way. Any ideas?
This is what I ended up doing based on #sammywemmy's suggestions
#Takes columns names and changes them into a list
names = df1.colnames.tolist()
#Merge df1 and df2 based on 'Item_ID'
merged = df1.merge(df2, on='Item_ID', how='outer')
for i in range(len(names)):
#using assign and **, we can bring in variable names with assign.
#Then add our Value 5 column
merged = merged.assign(**{names[i] : lambda x : x[names[i]] + x.Value5})
#Only keep all the columns before and including 'Value4'
df1= merged.loc[:,:'Value4']
Try this:
#set 'Item ID' as the index
df1 = df1.set_index('Item ID')
df2 = df2.set_index('Item ID')
#create list of columns that you are interested in
list_of_cols = ['Value1','Value2']
#create two separate dataframes
#unselected will not contain the columns you want to add
unselected = df1.drop(list_of_cols,axis=1)
#this will contain the columns you wish to add
selected = df1.filter(list_of_cols)
#reindex df2 so it has the same indices as df1
#then convert to a series
#fill the null values with 0
A = df2.reindex(index=selected.index,fill_value=0).loc[:,'Value5']
#add the series A to selected
selected = selected.add(A,axis='index')
#combine selected and unselected into one dataframe
result = pd.concat([unselected,selected],axis=1)
#this part is extra to get ur dataframe back to the way it was
#assumption here is that it is value1, value 2, bla bla
#so 1>2>3
#if ur columns are not actually Value1, Value2,
#bla bla, then a different sorting has to be used
#alternatively before the calculations,
#you could create a mapping of the columns to numbers
#that will give u a sorting mechanism and
#restore ur dataframe after calculations are complete
columns = sorted(result.columns,key = lambda x : x[-1])
#reindex back to the way it was
result = result.reindex(columns,axis='columns')
print(result)
Value1 Value2 Value3 Value4
Item ID
A 5 8 3 4
B 2 5 8 5
C 8 6 1 7
D 11 15 2 9
E 0 7 0 4
Alternative solution, using python's built-in dictionaries:
#create dictionaries
dict1 = (df1
#create temporary column
#and set as index
.assign(temp=df1['Item ID'])
.set_index('temp')
.to_dict('index')
)
dict2 = (df2
.assign(temp=df2['Item ID'])
.set_index('temp')
.to_dict('index')
)
list_of_cols = ['Value1','Value2']
intersected_keys = dict1.keys() & dict2.keys()
key_value_pair = [(key,col) for key in intersected_keys
for col in list_of_cols ]
#check for keys that are in both dict1 and 2
#loop through dict 1 and add values from dict2
#can be optimized with a dict comprehension
#leaving as is for better clarity IMHO
for key, val in key_value_pair:
dict1[key][val] = dict1[key][val] + dict2[key]['Value5']
#print(dict1)
{'A': {'Item ID': 'A', 'Value1': 5, 'Value2': 8, 'Value3': 3, 'Value4': 4},
'B': {'Item ID': 'B', 'Value1': 2, 'Value2': 5, 'Value3': 8, 'Value4': 5},
'C': {'Item ID': 'C', 'Value1': 8, 'Value2': 6, 'Value3': 1, 'Value4': 7},
'D': {'Item ID': 'D', 'Value1': 11, 'Value2': 15, 'Value3': 2, 'Value4': 9},
'E': {'Item ID': 'E', 'Value1': 0, 'Value2': 7, 'Value3': 0, 'Value4': 4}}
#create dataframe
pd.DataFrame.from_dict(dict1,orient='index').reset_index(drop=True)
Item ID Value1 Value2 Value3 Value4
0 A 5 8 3 4
1 B 2 5 8 5
2 C 8 6 1 7
3 D 11 15 2 9
4 E 0 7 0 4
I have a dataframe like below:
>>> df1
a b
0 [1, 2, 3] 10
1 [4, 5, 6] 20
2 [7, 8] 30
and another like:
>>> df2
a
0 1
1 2
2 3
3 4
4 5
I need to create column 'c' in df2 from column 'b' of df1 if column 'a' value of df2 is in coulmn 'a' df1. In df1 each tuple of column 'a' is a list.
I have tried to implement from following url, but got nothing so far:
https://medium.com/#Imaadmkhan1/using-pandas-to-create-a-conditional-column-by-selecting-multiple-columns-in-two-different-b50886fabb7d
expect result is
>>> df2
a c
0 1 10
1 2 10
2 3 10
3 4 20
4 5 20
Use Series.map by flattening values from df1 to dictionary:
d = {c: b for a, b in zip(df1['a'], df1['b']) for c in a}
print (d)
{1: 10, 2: 10, 3: 10, 4: 20, 5: 20, 6: 20, 7: 30, 8: 30}
df2['new'] = df2['a'].map(d)
print (df2)
a new
0 1 10
1 2 10
2 3 10
3 4 20
4 5 20
EDIT: I think problem is mixed integers in list in column a, solution is use if/else for test it for new dictionary:
d = {}
for a, b in zip(df1['a'], df1['b']):
if isinstance(a, list):
for c in a:
d[c] = b
else:
d[a] = b
df2['new'] = df2['a'].map(d)
Use :
m=pd.DataFrame({'a':np.concatenate(df.a.values),'b':df.b.repeat(df.a.str.len())})
df2.merge(m,on='a')
a b
0 1 10
1 2 10
2 3 10
3 4 20
4 5 20
First we unnest the list df1 to rows, then we merge them on column a:
df1 = df1.set_index('b').a.apply(pd.Series).stack().reset_index(level=0).rename(columns={0:'a'})
print(df1, '\n')
df_final = df2.merge(df1, on='a')
print(df_final)
b a
0 10 1.0
1 10 2.0
2 10 3.0
0 20 4.0
1 20 5.0
2 20 6.0
0 30 7.0
1 30 8.0
a b
0 1 10
1 2 10
2 3 10
3 4 20
4 5 20
I have df that looks like this:
a b c d e f
1 na 2 3 4 5
1 na 2 3 4 5
1 na 2 3 4 5
1 6 2 3 4 5
How do I trim and reshape the dataframe so that for every column the n/a are dropped and the dataframe looks like this:
Edit;
df.dropna() is dropping all the rows.
a b c d e f
1 6 2 3 4 5
This dataframe has millions of rows, I need to be able to drop the n/a rows by column while retaining rows and columns with data in them.
edit;
df.dropna() is dropping all the rows in the column. When I check if the columns with n/a are empty, df.column_name.empty() I get false. So there is data in columns with n/a
For me dropna working nice for remove missing values and Nones:
df = df.dropna()
print (df)
a b c d e f
3 1 6.0 2 3 4 5
But if possible multiple values for removing create mask by isin, chain testing missing values with isnull and last filter by any - return at least one True per row by inverted mask ~:
df = pd.DataFrame({'a': ['a', None, 's', 'd'],
'b': ['na',7, 2, 6],
'c': [2, 2, 2, 2],
'd': [3, 3, 3, 3],
'e': [4, 4, np.nan, 4],
'f': [5, 5, 5, 5]})
print (df)
a b c d e f
0 a na 2 3 4.0 5
1 None 7 2 3 4.0 5
2 s 2 2 3 NaN 5
3 d 6 2 3 4.0 5
df1 = df.dropna()
print (df1)
a b c d e f
0 a na 2 3 4.0 5
3 d 6 2 3 4.0 5
mask = (df.isin(['na', 'n/a']) | df.isnull()).any(axis=1)
df2 = df[~mask]
print (df2)
a b c d e f
3 d 6 2 3 4.0 5