I have a multi-indexed dataframe,
>>> df
a1 a2
b1 b2 b1 b2
c1 d1 11 21 31 41
d2 12 22 32 42
c2 d1 13 23 33 43
d2 14 24 34 44
It has 2 levels of header and 2 levels of index. If I directly use the code df.to_csv('test_file.csv'), then the format of the file test_file.csv is
,,a1,a1,a2,a2
,,b1,b2,b1,b2
c1,d1,11,21,31,41
c1,d2,12,22,32,42
c2,d1,13,23,33,43
c2,d2,14,24,34,44
However, I would like to change it to
remove the duplicates in the 1st level of header
remove entire 1st level of index, and make an empty row for each one in the 1st level of index.
The wanted format is:
,a1,,a2,
,b1,b2,b1,b2
c1,,,,,
d1,11,21,31,41
d2,12,22,32,42
c2,,,,,
d1,13,23,33,43
d2,14,24,34,44
Could you please show me how to do it? Thanks!
Please use the code below.
import pandas as pd
df = pd.DataFrame(
{
('a1', 'b1'): [11, 12, 13, 14],
('a1', 'b2'): [21, 22, 23, 24],
('a2', 'b1'): [31, 32, 33, 34],
('a2', 'b2'): [41, 42, 43, 44],
},
index=pd.MultiIndex.from_tuples([
('c1', 'd1'),
('c1', 'd2'),
('c2', 'd1'),
('c2', 'd2'),
]),
)
print(df)
df.to_csv('my_test_file.csv')
Here is a working solution. It uses a helper function to remove the duplicated consecutive labels and groupy+apply+pandas.concat to shift the multiindex level as extra empty row:
def remove_consecutive(l):
'''replaces consecutive items in "l" with empty string'''
from itertools import groupby, chain
return tuple(chain(*([k]+['']*(len(list(g))-1) for k,g in groupby(l))))
(df.groupby(level=0)
# below to shift level=0 as new row
.apply(lambda g: pd.concat([pd.DataFrame([],
index=[g.name],
columns=g.columns),
g.droplevel(0)]))
.droplevel(0)
# below to remove the duplicate column names
.T # unfortunately there is no set_index equivalent for columns, so transpose before/after
.set_index(pd.MultiIndex.from_arrays(list(map(remove_consecutive, zip(*df.columns)))))
.T
# export
.to_csv('filename.csv')
)
output:
,a1,,a2,
,b1,b2,b1,b2
c1,,,,
d1,11,21,31,41
d2,12,22,32,42
c1,,,,
d1,13,23,33,43
d2,14,24,34,44
Related
Im trying to move the f1_am, f2_am, f3_am to the correspondent column based on the values of f1_ty, f2_ty, f3_ty
I started adding new columns to the dataframe based on unique values from the _ty using sets, but I'm trying to figure it out how to move the _am values to were it belongs
Looked for the option of group by and pivot but the result exploded my mind....
I would appreciate some guidance.
Below the code.
import pandas as pd
import numpy as np
data = {
'mem_id': ['A', 'B', 'C', 'A', 'B', 'C']
, 'date_inf': ['01/01/2019', '01/01/2019', '01/01/2019', '02/01/2019', '02/01/2019', '02/01/2019']
, 'f1_ty': ['ABC', 'ABC', 'ABC', 'ABC', 'GHI', 'GHI']
, 'f1_am': [100, 20, 57, 44, 15, 10]
, 'f2_ty': ['DEF', 'DEF', 'DEF', 'GHI', 'ABC', 'XYZ']
, 'f2_am':[20, 30, 45, 66, 14, 21]
, 'f3_ty': ['XYZ', 'GHI', 'OPQ', 'OPQ', 'XYZ', 'DEF']
, 'f3_am':[20, 30, 45, 66, 14, 21]
}
df = pd.DataFrame (data)
#distinct values in columns using sets
distinct_values = sorted(list(set(df['f1_ty'])|set(df['f2_ty'])|set(df['f3_ty'])))
# add distinct values as new columns in the DataFrame
new_df = df.reindex(columns = np.append( df.columns.values, distinct_values))
So this would be my starting point and my wanted result.
Here is a try, thanks for the interesting problem (rename colujmns to make compatible to wide_to_long() followed by unstack() while dropping extra levels:
m=df.set_index(['mem_id','date_inf']).rename(columns=lambda x: ''.join(x.split('_')[::-1]))
n=(pd.wide_to_long(m.reset_index(),['tyf','amf'],['mem_id','date_inf'],'v')
.droplevel(-1).set_index('tyf',append=True).unstack(fill_value=0).reindex(m.index))
final=n.droplevel(0,axis=1).rename_axis(None,axis=1).reset_index()
print(final)
mem_id date_inf ABC DEF GHI OPQ XYZ
0 A 01/01/2019 100 20 0 0 20
1 B 01/01/2019 20 30 30 0 0
2 C 01/01/2019 57 45 0 45 0
3 A 02/01/2019 44 0 66 66 0
4 B 02/01/2019 14 0 15 0 14
5 C 02/01/2019 0 21 10 0 21
I want to sort a column of pandas DataFrame.
not just sorting but want to return a dataframe with sorted column.
in a sorted and updated DataFrame I want to concat a single column DataFrame.
students = {'name': ['s1', 's2', 's3', 's4', 's5', 's6', 's7', 's8', 's9', 's10']}
marks = {'grade': [45, 78, 12, 14, 48, 43, 47, 98, 35, 80]}
df_1 = pd.DataFrame(students)
df_2 = pd.DataFrame(marks)
df = pd.concat([df_1, df_2], axis=1)
df_25to_75 = df
df_25to_75.sort_values(['Marks'], inplace = True)
lst = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
a = pd.concat([df_25to_75, pd.DataFrame({'no.s': lst})], axis = 1)
You can create your DataFrame more simply, then you want rank. Feel free to sort in the end if it's necessary, though it's not to get the ranking.
import pandas as pd
df = pd.DataFrame({**students, **marks})
df['no.s'] = df.grade.rank(method='dense').astype(int)
name grade no.s
0 s1 45 5
1 s2 78 8
2 s3 12 1
3 s4 14 2
4 s5 48 7
5 s6 43 4
6 s7 47 6
7 s8 98 10
8 s9 35 3
9 s10 80 9
Your original issue is that though you sort the DataFrame the original index remains bound to the rows. Thus, when you then assign a new Series 1 - 10, it aligns on the original index, not on the row ordering.
I am using statsmodels.api on my data set. I have a list of panda series. The panda series has key value pairs. The keys are the names of the columns and the values contain the data. But, I have a list of series where the keys (column names) are repeated. I want to save all of the values from the list of pandas series to a single dataframe where the column names are the keys of the panda series. All of the series in the list have the same keys. I want to save them as a single data frame so that I can export the dataframe as a CSV. Any idea how I can save the keys as my column names of the df and then have the values fill the rest of the information.
Each series in the list returns something like this:
index 0 of the list: <class 'pandas.core.series.Series'>
height 23
weight 10
size 45
amount 9
index 1 of the list: <class 'pandas.core.series.Series'>
height 11
weight 99
size 25
amount 410
index 2 of the list: <class 'pandas.core.series.Series'>
height 3
weight 0
size 115
amount 92
I would like to be able to read a dataframe such that these values are saved as the following:
DataFrame:
height weight size amount
23 10 45 9
11 11 25 410
3 3 115 92
pd.DataFrame(data=your_list_of_series)
When creating a new DataFrame, pandas will accept a list of series for the data argument. The indices of your series will become the column names of the DataFrame.
Not the most efficient way, but this does the trick:
import pandas as pd
series_list =[ pd.Series({ 'height': 23,
'weight': 10,
'size': 45,
'amount': 9
}),
pd.Series({ 'height': 11,
'weight': 99,
'size': 25,
'amount': 410
}),
pd.Series({ 'height': 3,
'weight': 0,
'size': 115,
'amount': 92
})
]
pd.DataFrame( [series.to_dict() for series in series_list] )
Did you try just calling pd.DataFrame() on the list of series? That should just work.
import pandas as pd
series_list = [
pd.Series({
'height': 23,
'weight': 10,
'size': 45,
'amount': 9
}),
pd.Series({
'height': 11,
'weight': 99,
'size': 25,
'amount': 410
}),
pd.Series({
'height': 3,
'weight': 0,
'size': 115,
'amount': 92
})
]
df = pd.DataFrame(series_list)
print(df)
df.to_csv('path/to/save/foo.csv')
Output:
height weight size amount
0 23 10 45 9
1 11 99 25 410
2 3 0 115 92
I have a dataframe created by extracting data from a source (network wireless controller).
Dataframe is created off of a dictionary I build. This is basically what I am doing (a sample to show structure - not the actual dataframe):
df = pd.DataFrame({'AP-1': [30, 32, 34, 31, 33, 35, 36, 38, 37],
'AP-2': [30, 32, 34, 80, 33, 35, 36, 38, 37],
'AP-3': [30, 32, 81, 31, 33, 101, 36, 38, 37],
'AP-4': [30, 32, 34, 95, 33, 35, 103, 38, 121],
'AP-5': [30, 32, 34, 31, 33, 144, 36, 38, 37],
'AP-6': [30, 32, 34, 31, 33, 35, 36, 110, 37],
'AP-7': [30, 87, 34, 31, 111, 35, 36, 38, 122],
'AP-8': [30, 32, 99, 31, 33, 35, 36, 38, 37],
'AP-9': [30, 32, 34, 31, 33, 99, 88, 38, 37]}, index=['1', '2', '3', '4', '5', '6', '7', '8', '9'])
df1 = df.transpose()
This works fine.
Note about the data. Columns 1,2,3 are 'related'. They go together. Same for columns 4,5,6 and 7,8,9. I will explain more shortly.
Columns 1, 4, 7 are client count. Columns 2, 5, 8 are channel util on the 5 Ghz spectrum. Columns 3, 6, 9 are channel util on the 2.4 Ghz spectrum.
Basically I take a reading at 5 minute intervals. The above would represent three readings at 5 minute intervals.
What I want is two new dataframes, two columns each, constructed as follows:
Examine the 5 Ghz columns (here it is 2, 5, 8). Which ever has the highest value becomes column 1 in the new dataframe. Column 2 would be the value of the client count column related to the 5 Ghz column with the highest value. In other words, if column 2 were the highest out of columns 2, 5, 8, then I want the value in column 1 to be the value in the new dataframe for the second column. If the value in column 8 were highest, then I want to also pull the value in column 7. I want the index to be same in the new dataframes as the original -- AP name.
I want to do this for all rows in the 'main' dataframe. I want two new dataframes -- so I will repeat this exact procedure for the 5 Ghz columns and the 2.4 (columns 3, 6, 9 -- also grabbing the corresponding highest client count value for the second column in the new dataframe.
What I have tried:
First I broke the main dataframe into three: df1 has all the client count columns, df2 has the 5 Ghz, and df3 has the 2.4 info, using this:
# create client count only dataframe
df_cc = df[df.columns[::3]]
print(df_cc)
print()
# create 5Ghz channel utilization only dataframe
df_5Ghz = df[df.columns[1::3]]
print(df_5Ghz)
print()
# create 2.4Ghz channel utilization only dataframe
df_24Ghz = df[df.columns[2::3]]
print(df_24Ghz)
print()
This works.
I thought I could then reference the main dataframe, but I don't know how.
Then I found this:
extract column value based on another column pandas dataframe
The query option looked great, but I don't know the value. I need to first discover the max value of the 2.4 and 5 Ghz columns respectively, then grab the corresponding client count value. That is why I first created dataframes containing the 2.4 and 5 Ghz values only, thinking I could first get the max value of each row, then do a lookup on the main dataframe (or use the client count onlydataframe I created), but I just do not know how to realize this idea.
Any assistance would be greatly appreciated.
You can get what you want in 3 steps:
# connection between columns
mapping = {'2': '1', '5': '4', '8': '7'}
# 1. column with highest value among 5GHz values (pandas series)
df2 = df1.loc[:, ['2', '5', '8']].idxmax(axis=1)
df2.name = 'highest value'
# 2. column with client count corresponding to the highest value (pandas series)
df3 = df2.apply(lambda x: mapping[x])
df3.name = 'client count'
# 3. build result using 2 lists of columns (pandas dataframe)
df4 = pd.DataFrame(
{df.name: [
df1.loc[idx, col]
for idx, col in zip(df.index, df.values)]
for df in [df2, df3]},
index=df1.index)
print(df4)
Output:
highest value client count
AP-1 38 36
AP-2 38 36
AP-3 38 36
AP-4 38 103
AP-5 38 36
AP-6 110 36
AP-7 111 31
AP-8 38 36
AP-9 38 88
I guess while not sure it would be easier to solve the issue (and faster to compute) without pandas using just built-in python data types - dictionaries and lists.
I would like to find a better way to carry out the following process.
#import packages
import pandas as pd
I have defined a pandas dataframe.
# Create dataframe
data = {'name': ['Jason', 'Jason', 'Tina', 'Tina', 'Tina'],
'reports': [4, 24, 31, 2, 3],
'coverage': [25, 94, 57, 62, 70]}
df = pd.DataFrame(data)
After the dataframe is created, I want to add an extra column to the dataframe. This column contains the rank based on the values in the coverage column for every name seperately.
#Add column with ranks based on 'coverage' for every name separately.
df_end = pd.DataFrame()
for person_names in df.groupby('name').groups:
one_name = df.groupby('name').get_group(person_names)
one_name['coverageRank'] = one_name['coverage'].rank()
df_end = df_end.append(one_name)
Is it possible to achieve this simple task in a simpler way? Maybe without using the for-loop?
I think you need DataFrameGroupBy.rank:
df['coverageRank'] = df.groupby('name')['coverage'].rank()
print (df)
coverage name reports coverageRank
0 25 Jason 4 1.0
1 94 Jason 24 2.0
2 57 Tina 31 1.0
3 62 Tina 2 2.0
4 70 Tina 3 3.0