I have a dataframe with three columns and a function that calculates the values of column y and z given the value of column x. I need to only calculate the values if they are missing NaN.
def calculate(x):
return 1, 2
df = pd.DataFrame({'x':['a', 'b', 'c', 'd', 'e', 'f'], 'y':[np.NaN, np.NaN, np.NaN, 'a1', 'b2', 'c3'], 'z':[np.NaN, np.NaN, np.NaN, 'a2', 'b1', 'c4']})
x y z
0 a NaN NaN
1 b NaN NaN
2 c NaN NaN
3 d a1 a2
4 e b2 b1
5 f c3 c4
mask = (df.isnull().any(axis=1))
df[['y', 'z']] = df[mask].apply(calculate, axis=1, result_type='expand')
However, I get the following result, although I only apply to the masked set. Unsure what I'm doing wrong.
x y z
0 a 1.0 2.0
1 b 1.0 2.0
2 c 1.0 2.0
3 d NaN NaN
4 e NaN NaN
5 f NaN NaN
If the mask is inverted I get the following result:
df[['y', 'z']] = df[~mask].apply(calculate, axis=1, result_type='expand')
x y z
0 a NaN NaN
1 b NaN NaN
2 c NaN NaN
3 d 1.0 2.0
4 e 1.0 2.0
5 f 1.0 2.0
Expected result:
x y z
0 a 1.0 2.0
1 b 1.0 2.0
2 c 1.0 2.0
3 d a1 a2
4 e b2 b1
5 f c3 c4
you can fillna after calculating for the full dataframe and set_axis
out = (df.fillna(df.apply(calculate, axis=1, result_type='expand')
.set_axis(['y','z'],inplace=False,axis=1)))
print(out)
x y z
0 a 1 2
1 b 1 2
2 c 1 2
3 d a1 a2
4 e b2 b1
5 f c3 c4
Try:
df.loc[mask,["y","z"]] = pd.DataFrame(df.loc[mask].apply(calculate, axis=1).to_list(), index=df[mask].index, columns = ["y","z"])
print(df)
x y z
0 a 1 2
1 b 1 2
2 c 1 2
3 d a1 a2
4 e b2 b1
5 f c3 c4
Related
I have a Pandas dataframe with the following structure:
A B C
a b 1
a b 2
a b 3
c d 7
c d 8
c d 5
c d 6
c d 3
e b 4
e b 3
e b 2
e b 1
And I will like to transform it into this:
A B C1 C2 C3 C4 C5
a b 1 2 3 NAN NAN
c d 7 8 5 6 3
e b 4 3 2 1 NAN
In other words, something like groupby A and B and expand C into different columns.
Knowing that the length of each group is different.
C is already ordered
Shorter groups can have NAN or NULL values (empty), it does not matter.
Use GroupBy.cumcount and pandas.Series.add with 1, to start naming the new columns from 1 onwards, then pass this to DataFrame.pivot, and add DataFrame.add_prefix to rename the columns (C1, C2, C3, etc...). Finally use DataFrame.rename_axis to remove the indexes original name ('g') and transform the MultiIndex into columns by using DataFrame.reset_indexcolumns A,B:
df['g'] = df.groupby(['A','B']).cumcount().add(1)
df = df.pivot(['A','B'], 'g', 'C').add_prefix('C').rename_axis(columns=None).reset_index()
print (df)
A B C1 C2 C3 C4 C5
0 a b 1.0 2.0 3.0 NaN NaN
1 c d 7.0 8.0 5.0 6.0 3.0
2 e b 4.0 3.0 2.0 1.0 NaN
Because NaN is by default of type float, if you need the columns dtype to be integers add DataFrame.astype with Int64:
df['g'] = df.groupby(['A','B']).cumcount().add(1)
df = (df.pivot(['A','B'], 'g', 'C')
.add_prefix('C')
.astype('Int64')
.rename_axis(columns=None)
.reset_index())
print (df)
A B C1 C2 C3 C4 C5
0 a b 1 2 3 <NA> <NA>
1 c d 7 8 5 6 3
2 e b 4 3 2 1 <NA>
EDIT: If there's a maximum N new columns to be added, it means that A,B are duplicated. Therefore, it will beneeded to add helper groups g1, g2 with integer and modulo division, adding a new level in index:
N = 4
g = df.groupby(['A','B']).cumcount()
df['g1'], df['g2'] = g // N, (g % N) + 1
df = (df.pivot(['A','B','g1'], 'g2', 'C')
.add_prefix('C')
.droplevel(-1)
.rename_axis(columns=None)
.reset_index())
print (df)
A B C1 C2 C3 C4
0 a b 1.0 2.0 3.0 NaN
1 c d 7.0 8.0 5.0 6.0
2 c d 3.0 NaN NaN NaN
3 e b 4.0 3.0 2.0 1.0
df1.astype({'C':str}).groupby([*'AB'])\
.agg(','.join).C.str.split(',',expand=True)\
.add_prefix('C').reset_index()
A B C0 C1 C2 C3 C4
0 a b 1 2 3 None None
1 c d 7 8 5 6 3
2 e b 4 3 2 1 None
The accepted solution but avoiding the deprecation warning:
N = 3
g = df_grouped.groupby(['A','B']).cumcount()
df_grouped['g1'], df_grouped['g2'] = g // N, (g % N) + 1
df_grouped = (df_grouped.pivot(index=['A','B','g1'], columns='g2', values='C')
.add_prefix('C_')
.astype('Int64')
.droplevel(-1)
.rename_axis(columns=None)
.reset_index())
I have 2 dfs, which I want to combine as the following:
df1 = pd.DataFrame({"a": [1,2], "b":['A','B'], "c":[3,2]})
df2 = pd.DataFrame({"a": [1,1,1, 2,2,2, 3, 4], "b":['A','A','A','B','B', 'B','C','D'], "c":[3, None,None,2,None,None,None,None]})
Output:
a b c
1 A 3.0
1 A NaN
1 A NaN
2 B 2.0
2 B NaN
2 B NaN
I had an earlier version of this question that only involved df2 and was solved with
df.groupby(['a','b']).filter(lambda g: any(~g['c'].isna()))
but now I need to run it only for rows that appear in df1 (df2 contains rows from df1 but some extra rows which I want to not be included.
Thanks!
You can turn the indicator on with merge
out = df2.merge(df1,indicator=True,how='outer',on=['a','b'])
Out[91]:
a b c_x c_y _merge
0 1 A 3.0 3.0 both
1 1 A NaN 3.0 both
2 1 A NaN 3.0 both
3 2 B 2.0 2.0 both
4 2 B NaN 2.0 both
5 2 B NaN 2.0 both
6 3 C NaN NaN left_only
7 4 D NaN NaN left_only
out = out[out['_merge']=='both']
IIUC, you could merge:
out = df2.merge(df1[['a','b']])
or you could use chained isin:
out1 = df2[df2['a'].isin(df1['a']) & df2['b'].isin(df1['b'])]
Output:
a b c
0 1 A 3.0
1 1 A NaN
2 1 A NaN
3 2 B 2.0
4 2 B NaN
5 2 B NaN
I have three dataframes df1, df2, and df3, which are defined as follows
df1 =
A B C
0 1 a a1
1 2 b b2
2 3 c c3
3 4 d d4
4 5 e e5
5 6 f f6
df2 =
A B C
0 1 a X
1 2 b Y
2 3 c Z
df3 =
A B C
3 4 d P
4 5 e Q
5 6 f R
I have defined a Primary Key list PK = ["A","B"].
Now, I take a fourth dataframe df4 as df4 = df1.sample(n=2), which gives something like
df4 =
A B C
4 5 e e5
1 2 b b2
Now, I want to select the rows from df2 and df1 which matches the values of the primary keys of df4.
For eg, in this case,
I need to get row with
index = 4 from df3,
index = 1 from df2.
If possible I need to get a dataframe as follows:
df =
A B C A(df2) B(df2) C(df2) A(df3) B(df3) C(df3)
4 5 e e5 5 e Q
1 2 b b2 2 b Y
Any ideas on how to work this out will be very helpful.
Use two consecutive DataFrame.merge operations along with using DataFrame.add_suffix on the right dataframe to left merge the dataframes df4, df2, df3, finally use Series.fillna to replace the missing values with empty string:
df = (
df4.merge(df2.add_suffix('(df2)'), left_on=['A', 'B'], right_on=['A(df2)', 'B(df2)'], how='left')
.merge(df3.add_suffix('(df3)'), left_on=['A', 'B'], right_on=['A(df3)', 'B(df3)'], how='left')
.fillna('')
)
Result:
# print(df)
A B C A(df2) B(df2) C(df2) A(df3) B(df3) C(df3)
0 5 e e5 5 e Q
1 2 b b2 2 b Y
Here's how I would do it on the entire data set. If you want to sample first, just update the merge statements at the end by replacing df1 with df4 or just take a sample of t
PK = ["A","B"]
df2 = pd.concat([df2,df2], axis=1)
df2.columns=['A','B','C','A(df2)', 'B(df2)', 'C(df2)']
df2.drop(columns=['C'], inplace=True)
df3 = pd.concat([df3,df3], axis=1)
df3.columns=['A','B','C','A(df3)', 'B(df3)', 'C(df3)']
df3.drop(columns=['C'], inplace=True)
t = df1.merge(df2, on=PK, how='left')
t = t.merge(df3, on=PK, how='left')
Output
A B C A(df2) B(df2) C(df2) A(df3) B(df3) C(df3)
0 1 a a1 1.0 a X NaN NaN NaN
1 2 b b2 2.0 b Y NaN NaN NaN
2 3 c c3 3.0 c Z NaN NaN NaN
3 4 d d4 NaN NaN NaN 4.0 d P
4 5 e e5 NaN NaN NaN 5.0 e Q
5 6 f f6 NaN NaN NaN 6.0 f R
I have DataFrame x and I would like to replace one column with Series y
x = DataFrame([[1,2],[3,4]], columns=['C1','C2'], index=['a','b'])
C1 C2
a 1 2
b 3 4
y = Series([5,6,7], index=['a','b','c'])
a 5
b 6
c 7
Simple replacement works fine but keeps original index of DataFrame
x['C1'] = y
C1 C2
a 5 2
b 6 4
I need to have union of indeces of x and y. One solution would be to reindex before replacement
x = x.reindex(x.index.union(y.index), copy=False)
x['C1'] = y
C1 C2
a 5 2.0
b 6 4.0
c 7 NaN
Is there simpler way?
combine_first
Turn y into a DataFrame first with to_frame
y.to_frame('C1').combine_first(x)
C1 C2
a 5 2.0
b 6 4.0
c 7 NaN
align and assign
Use align to... align the indices
x, y = x.align(y, axis=0)
x.assign(C1=y)
C1 C2
a 5 2.0
b 6 4.0
c 7 NaN
Your cat try use join:
pd.DataFrame(y,columns=['C1']).join(x[['C2']])
Output:
C1 C2
a 5 2.0
b 6 4.0
c 7 NaN
Similar to your solution but more succinct, you use reindex, then assign:
res = x.reindex(x.index | y.index).assign(C1=y)
print(res)
C1 C2
a 5 2.0
b 6 4.0
c 7 NaN
You can use concat but you will have to fix the column names, i.e.
import pandas as pd
pd.concat([x.loc[:, 'C2'], y], axis = 1)
which gives,
C2 0
a 2.0 5
b 4.0 6
c NaN 7
I am trying to unstack two columns :
cols = res.columns[:31]
res[cols] = res[cols].ffill()
res = res.set_index(cols + [31])[32].unstack().reset_index().rename_axis(None, 1)
But I am getting an error :
TypeError: can only perform ops with scalar values
What should I do to avoid it?
My original problem : LINK
I think need convert columns to list:
cols = res.columns[:31].tolist()
EDIT:
Index contains duplicate entries, cannot reshape
means duplicates, here for first 6 columns, so is impossible create new DataFrame, because first 6 column create new index and 7. column create new column, and for 8. column are 2 values:
0 1 2 3 4 5 6 7
0 xx s 1 d f df f 54
1 xx s 1 d f df f g4
New DataFrame:
index = xx s 1 d f df
column = f
value = 54
index = xx s 1 d f df
column = f
value = g4
So solution is aggregate, here working with strings, so need .apply(', '.join):
index = xx s 1 d f df
column = f
value = 54, g4
Or remove duplicates and keep first or last value of dupes rows by drop_duplicates:
index = xx s 1 d f df
column = f
value = 54
index = xx s 1 d f df
column = f
value = g4
res = pd.DataFrame({0: ['xx',np.nan,np.nan,np.nan,'ds', np.nan, np.nan, np.nan, np.nan, 'as'],
1: ['s',np.nan,np.nan,np.nan,'a', np.nan, np.nan, np.nan, np.nan, 't'],
2: ['1',np.nan,np.nan,np.nan,'s', np.nan, np.nan, np.nan, np.nan, 'r'],
3: ['d',np.nan, np.nan, np.nan,'d', np.nan, np.nan, np.nan, np.nan, 'a'],
4: ['f',np.nan, np.nan, np.nan,'f', np.nan, np.nan, np.nan, np.nan, '2'],
5: ['df',np.nan,np.nan,np.nan,'ds',np.nan, np.nan, np.nan, np.nan, 'ds'],
6: ['f','f', 'x', 'r', 'f', 'd', 's', '1', '3', 'k'],
7: ['54','g4', 'r4', '43', '64', '43', 'se', 'gf', 's3', 's4']})
cols = res.columns[:6].tolist()
res[cols] = res[cols].ffill()
print (res)
0 1 2 3 4 5 6 7
0 xx s 1 d f df f 54
1 xx s 1 d f df f g4
2 xx s 1 d f df x r4
3 xx s 1 d f df r 43
4 ds a s d f ds f 64
5 ds a s d f ds d 43
6 ds a s d f ds s se
7 ds a s d f ds 1 gf
8 ds a s d f ds 3 s3
9 as t r a 2 ds k s4
res =res.groupby(cols + [6])[7].apply(', '.join).unstack().reset_index().rename_axis(None, 1)
print (res)
0 1 2 3 4 5 1 3 d f k r s x
0 as t r a 2 ds NaN NaN NaN NaN s4 NaN NaN NaN
1 ds a s d f ds gf s3 43 64 NaN NaN se NaN
2 xx s 1 d f df NaN NaN NaN 54, g4 NaN 43 NaN r4 <-54, g4
Another solution is remove duplicates:
res = res.drop_duplicates(cols + [6])
res = res.set_index(cols + [6])[7].unstack().reset_index().rename_axis(None, 1)
print (res)
0 1 2 3 4 5 1 3 d f k r s x
0 as t r a 2 ds NaN NaN NaN NaN s4 NaN NaN NaN
1 ds a s d f ds gf s3 43 64 NaN NaN se NaN
2 xx s 1 d f df NaN NaN NaN 54 NaN 43 NaN r4 <- 54
res = res.drop_duplicates(cols + [6], keep='last')
res = res.set_index(cols + [6])[7].unstack().reset_index().rename_axis(None, 1)
print (res)
0 1 2 3 4 5 1 3 d f k r s x
0 as t r a 2 ds NaN NaN NaN NaN s4 NaN NaN NaN
1 ds a s d f ds gf s3 43 64 NaN NaN se NaN
2 xx s 1 d f df NaN NaN NaN g4 NaN 43 NaN r4 <- g4