Data handling and manipulation with Pandas

Renesh Bedre    13 minute read

What is pandas?

Pandas is python data analysis library which provides integrated and function rich utilities for data handling, wrangling, and analysis of the datasets.

How to install pandas?

# using Miniconda
# to install specific version provide pandas=version number (e.g. pandas=1.0)
conda install pandas

# using pip
pip install pandas

Check pandas version

# you can use interactive python interpreter, jupyter notebook or python code
# I am using interactive python interpreter (Python 3.7)
>>> import pandas as pd
>>> pd.__version__
'0.25.3'

Create dataframe and series using pandas

# you can use interactive python interpreter, jupyter notebook or python code
# I am using interactive python interpreter (Python 3.7)

# create pandas dataframe (two-dimensional)
>>> import pandas as pd
# dataframe from dict
>>> df1 = pd.DataFrame({'col1':['A', 'B', 'C', 'D', 'E'], 'col2':[1,2,3,4,5], 'col3':[0.1,0.2,0.3,0.4,0.5]})
>>> df1
col1 	col2 	col3
0 	A 	1 	0.1
1 	B 	2 	0.2
2 	C 	3 	0.3
3 	D 	4 	0.4
4 	E 	5 	0.5

# Create a dataframe with some random numbers
>>> import pandas as pd
>>> import numpy as np
>>> df2 = pd.DataFrame(np.random.randint(0,100,size=(1000, 10)), columns=['col%i' % i for i in np.arange(10)])
# print first 5 rows
>>> df2.head()
   col0  col1  col2  col3  col4  col5  col6  col7  col8  col9
0     2    72     7    23    82    76    29    86    55    41
1    37    23    73    62    53    70    50    30     5    11
2    38    13    64    54    19    35    93    67    25    13
3    55    58    82    92    35    24     0    57     7     6
4    75    65    82    68    16    49    50    86    52    65

# create pandas series (one-dimensional)
df3 = pd.Series(np.random.randint(0,100, size=10))
>>> df3.head()
0     5
1    37
2    59
3    53
4    37
dtype: int32

Import data from file

# read csv file
>>> import pandas as pd
>>> df4 = pd.read_csv("https://reneshbedre.github.io/assets/posts/volcano/testvolcano.csv")
>>> df4.head()
          GeneNames  value1  value2    log2FC       p-value
0  LOC_Os09g01000.1    8862   32767 -1.886539  1.250000e-55
1  LOC_Os12g42876.1    1099     117  3.231611  1.050000e-55
2  LOC_Os12g42884.2     797      88  3.179004  2.590000e-54
3  LOC_Os03g16920.1     274       7  5.290677  4.690000e-54
4  LOC_Os05g47540.4     308      18  4.096862  2.190000e-54

# read tab file
>>> df5 = pd.read_csv("https://reneshbedre.github.io/assets/posts/anova/onewayanova.txt", sep="\t")
>>> df5.head()
    A   B   C   D
0  25  45  30  54
1  30  55  29  60
2  28  29  33  51
3  36  56  37  62
4  29  40  27  73

# make first column of table as index
>>> df6 = pd.read_csv("https://reneshbedre.github.io/assets/posts/volcano/testvolcano.csv", index_col=0)
>>> df6.head()
                  value1  value2    log2FC       p-value
GeneNames
LOC_Os09g01000.1    8862   32767 -1.886539  1.250000e-55
LOC_Os12g42876.1    1099     117  3.231611  1.050000e-55
LOC_Os12g42884.2     797      88  3.179004  2.590000e-54
LOC_Os03g16920.1     274       7  5.290677  4.690000e-54
LOC_Os05g47540.4     308      18  4.096862  2.190000e-54

# import specific columns only
# you can also replace numbers with column names in usecols argument
>>> df7 = pd.read_csv("https://reneshbedre.github.io/assets/posts/volcano/testvolcano.csv", usecols=[0,3,4])
>>> df7.head()
          GeneNames    log2FC       p-value
0  LOC_Os09g01000.1 -1.886539  1.250000e-55
1  LOC_Os12g42876.1  3.231611  1.050000e-55
2  LOC_Os12g42884.2  3.179004  2.590000e-54
3  LOC_Os03g16920.1  5.290677  4.690000e-54
4  LOC_Os05g47540.4  4.096862  2.190000e-54

# skip first n rows
# you can also skip rows from bottom using skipfooter argument
>>> df8 = pd.read_csv("https://reneshbedre.github.io/assets/posts/volcano/testvolcano.csv", index_col=0, skiprows=5, header=None)
>>> df8.head()
                    1     2         3             4
0
LOC_Os05g47540.4  308    18  4.096862  2.190000e-54
LOC_Os09g00999.1  339  1213 -1.839222  1.950000e-54
LOC_Os01g62900.1  193     4  5.592457  1.560000e-54
LOC_Os09g27030.2  504   162  1.637430  1.360000e-54
LOC_Os01g51410.2  225    22  3.354350  6.760000e-54

# read only first n rows
>>> df9 = pd.read_csv("https://reneshbedre.github.io/assets/posts/volcano/testvolcano.csv", index_col=0, nrows=100)
>>> df9
                  value1  value2    log2FC       p-value
GeneNames
LOC_Os09g01000.1    8862   32767 -1.886539  1.250000e-55
LOC_Os12g42876.1    1099     117  3.231611  1.050000e-55
LOC_Os12g42884.2     797      88  3.179004  2.590000e-54
LOC_Os03g16920.1     274       7  5.290677  4.690000e-54
LOC_Os05g47540.4     308      18  4.096862  2.190000e-54
[100 rows x 4 columns]

# skip comment lines
# below command will skip any lines starting with #
# only one character allowed for comments (# is okay, but ## will not work)
>>> df = pd.read_csv("file.csv", comment='#')

Check data type (dtype)

Check a data type of each column in the dataframe

>>> df1.dtypes
Col1     object
col2      int64
col3    float64
dtype: object

Check dataframe dimensions (size, shape and ndim)

Check the total number of elements, columns, and rows in a dataframe

# get total number of elements in a dataframe 
>>> df1.size
15
>>> df2.size
10000

# get number of rows and columns
>>> df1.shape
(5, 3)  # (rows, columns)
>>> df2.shape
(1000, 10) # (rows, columns)

# get the array dimensions (number of axes)
# returns 1 for pandas series (one-dimensional) and 2 for dataframes (two-dimensional) 
>>> df1.ndim
2
>>> df2.ndim
2
>>> df3.ndim
1

Extracting columns and rows from dataframe

# get column data based on column name
>>> df1['col2']
0    1
1    2
2    3
3    4
4    5
Name: col2, dtype: int64

# get row data based on row index
# return row output will be series 
# extract row with index 0 
>>> df1.loc[0] 
Col1      A
col2      1
col3    0.1
Name: 0, dtype: object

# assign first column (col1) as index
>>> df1_1 = df1.set_index('col1')
>>> df1_1
      col2  col3
col1
A        1   0.1
B        2   0.2
C        3   0.3
D        4   0.4
E        5   0.5

# get rows data based on row index (col1 as index)
>>> df1_1.loc['E']
col2    5.0
col3    0.5
Name: E, dtype: float64

Extracting subsets of dataframes

# get subset of dataframe [row, column]
>>> df1_1.loc['A':'C', 'col2':'col3']
      col2  col3
col1
A        1   0.1
B        2   0.2
C        3   0.3

# based on default numeric index
>>> df2.loc[5:9, 'col2':'col6']
   col2  col3  col4  col5  col6
5    49     5    61    87    23
6    20    57    40    21    17
7    30    30    38    94    50
8    98    17    31    67    59
9    39     9     4    17     8

# get specific columns with all rows
>>> df2[['col1', 'col5', 'col8']]
     col1  col5  col8
0      21     2    14
1      93    29    93
2      23    24    32
3      26    72    55
4      17    96    40

# get specific rows with all columns
>>> df2[2:6]
   col0  col1  col2  col3  col4  col5  col6  col7  col8  col9
2     4    23    64    49    71    24    97    92    32    39
3    73    26    33    99    62    72    73     3    55    82
4    72    17    73    71    42    96    56    29    40    96
5    16    10    49     5    61    87    23     7    36     1

# get specific columns with specific rows
>>> df2.loc[2:4, ['col1', 'col5', 'col8']]


Extracting rows and columns based on conditional match

import pandas as pd
# dataframe from dict
df = pd.DataFrame({'col1':['A', 'A', 'C', 'D', 'E'], 'col2':[1,2,3,4,5], 'col3':[0.1,0.2,0.3,0.4,0.5]})
df
# output
  col1  col2  col3
0    A     1   0.1
1    A     2   0.2
2    C     3   0.3
3    D     4   0.4
4    E     5   0.5

# extract rows where col1 has values A 
df.loc[df['col1'] == 'A']
# output
  col1  col2  col3
0    A     1   0.1
1    A     2   0.2

# extract rows where col1 has values A and D
df.loc[df['col1'].isin(['A', 'D'])]
# output
  col1  col2  col3
0    A     1   0.1
1    A     2   0.2
3    D     4   0.4

Statistics of dataframes

# check above to create dataframe objects (df1, df2 and df1_1)
# get descriptive statistics (count, mean, std dev, min, max, and quartiles)
# it excludes all NaN or missing values
>>> df1.describe()
           col2      col3
count  5.000000  5.000000
mean   3.000000  0.300000
std    1.581139  0.158114
min    1.000000  0.100000
25%    2.000000  0.200000
50%    3.000000  0.300000
75%    4.000000  0.400000
max    5.000000  0.500000

# get only certain element like mean ,count
# get mean
>>> df1.describe().loc['mean']
col2    3.0
col3    0.3
Name: mean, dtype: float64

# get mean and std dev
>>> df1.describe().loc[['mean', 'std']]
          col2      col3
mean  3.000000  0.300000
std   1.581139  0.158114

# get standard error of mean (sem)
>>> df1.sem()
col2    0.707107
col3    0.070711
dtype: float64

# for specific columns only
>>> df2['col4'].describe()
count    1000.000000
mean       48.917000
std        28.679865
min         0.000000
25%        25.000000
50%        49.000000
75%        74.000000
max        99.000000
Name: col4, dtype: float64

# for only selected multiple columns
>>> df2.loc[:,'col1':'col5'].describe()
              col1         col2         col3         col4         col5
count  1000.000000  1000.000000  1000.000000  1000.000000  1000.000000
mean     49.334000    51.057000    48.676000    48.917000    48.486000
std      28.311375    29.141293    28.252065    28.679865    29.070619
min       0.000000     0.000000     0.000000     0.000000     0.000000
25%      25.000000    26.000000    25.000000    25.000000    22.000000
50%      50.000000    52.000000    49.000000    49.000000    50.000000
75%      73.000000    77.000000    73.000000    74.000000    73.000000
max      99.000000    99.000000    99.000000    99.000000    99.000000

# to get descriptive statistics by row-wise you can transpose the columns
# or use apply function
>>> df1_1.apply(pd.DataFrame.describe, axis=1)
      count  mean       std  min    25%   50%    75%  max
col1
A       2.0  0.55  0.636396  0.1  0.325  0.55  0.775  1.0
B       2.0  1.10  1.272792  0.2  0.650  1.10  1.550  2.0
C       2.0  1.65  1.909188  0.3  0.975  1.65  2.325  3.0
D       2.0  2.20  2.545584  0.4  1.300  2.20  3.100  4.0
E       2.0  2.75  3.181981  0.5  1.625  2.75  3.875  5.0

# to get descriptive statistics by row-wise but only for specific columns
>> df2.loc[:,'col0':'col4'].apply(pd.DataFrame.describe, axis=1)
     count  mean        std   min   25%   50%   75%   max
0      5.0  43.8  37.325594  12.0  12.0  30.0  69.0  96.0
1      5.0  52.8  34.687173   1.0  43.0  55.0  71.0  94.0
2      5.0  61.6  36.287739  14.0  34.0  72.0  90.0  98.0
3      5.0  70.0  32.992423  21.0  52.0  83.0  96.0  98.0
4      5.0  43.4  34.681407   3.0   9.0  59.0  73.0  73.0
..

Add/Insert rows and columns to existing dataframes

# check above to create dataframe objects (df1, df2 and df1_1)
# add extra row to existing df1 dataframe
>>>df1_2 = df1.append({'col1': 'F', 'col2': 6, 'col3': 0.6}, ignore_index=True)
# ignore_index option will ignore the current index and append to df1
# dataframe.This option is useful when appending two dataframe and to
# preserve the original index of each dataframes
>>> df1_2
  col1  col2  col3
0    A     1   0.1
1    B     2   0.2
2    C     3   0.3
3    D     4   0.4
4    E     5   0.5
5    F     6   0.6

# adding a column to dataframe
>>> df1_2['col4'] = [1, 4, 9, 16, 25,36]
>>> df1_2
  col1  col2  col3  col4
0    A     1   0.1     1
1    B     2   0.2     4
2    C     3   0.3     9
3    D     4   0.4    16
4    E     5   0.5    25
5    F     6   0.6    36

# merge/append two dataframes
# create a new dataframe 
>>> df10 = pd.DataFrame({'col1':['X', 'Y'], 'col2':[7,8], 'col3':[0.7,0.8]})
>>> df10
  col1  col2  col3
0    X     7   0.7
1    Y     8   0.8

# mereg df10 with df1_2
>>> df1_2.append(df10)
  col1  col2  col3  col4
0    A     1   0.1   1.0
1    B     2   0.2   4.0
2    C     3   0.3   9.0
3    D     4   0.4  16.0
4    E     5   0.5  25.0
5    F     6   0.6  36.0
0    X     7   0.7   NaN
1    Y     8   0.8   NaN

# if you set ignore_index=True, the index order will not be preserved
>>> df1_2.append(df10, ignore_index=True)
  col1  col2  col3  col4
0    A     1   0.1   1.0
1    B     2   0.2   4.0
2    C     3   0.3   9.0
3    D     4   0.4  16.0
4    E     5   0.5  25.0
5    F     6   0.6  36.0
6    X     7   0.7   NaN
7    Y     8   0.8   NaN

Export dataframes to file

# check above to create dataframe objects (df1)
# Export to CSV file
# with first column as index
>>> df1.to_csv("df1.csv")
# do not export index as first column
>>> df1.to_csv("df1.csv", index=False)
# Export to TAB file
>>> df1.to_csv("df1.csv", sep='\t')
# no column headers
>>> df1.to_csv("df1.csv", header=False)
# compress output file (works with pandas 1.0.0)
>>> df1.to_csv("df1.zip", compression={'method':'zip', 'archive_name':'df1.csv'})
# export to excel file
# use similar option as above for headers and index
>>> df1.to_excel("df1.xlsx")

Convert series to dataframes

# create pandas series
>>> df11 = pd.Series([1,2,3])
>>> df11
0    1
1    2
2    3
dtype: int64

# convert to dataframe
>>> df11.to_frame()
   0
0  1
1  2
2  3

# add column name
>>> df11.to_frame('col1')
   col1
0     1
1     2
2     3

Convert numpy arrays to dataframes

>>> import numpy as np
>>> x = np.array([[ 1,  2], [ 3 ,4]])

# convert to dataframe
df12 = pd.DataFrame({'col1':x[:,0], 'col2':x[:,1]})
>>> df12
   col1  col2
0     1     2
1     3     4

Convert dataframe to numpy arrays

# see df12 dataframe in above section
# convert whole dataframe to numpy array
>>> df12.to_numpy()
array([[1, 2],
       [3, 4]])

# extract specific column as numpy array
>>> df12['col1'].to_numpy()
array([1, 3])

# extract multiple columns as numpy array
>>> df12[['col1','col2']].to_numpy()
array([[1, 2],
       [3, 4]])

# extract specific row as numpy array
>>> df12.loc[0].to_numpy()
array([1, 2])

# extract multiple row as numpy array
>>> df12.loc[0:1].to_numpy()
array([[1, 2],
       [3, 4]])

Find min and max values in dataframe

Create a dataframe with some random numbers

>>> import pandas as pd
>>> import numpy as np
>>> df = pd.read_csv("https://reneshbedre.github.io/assets/posts/anova/onewayanova.txt", sep="\t")
>>> df.head()
    A   B   C   D
0  25  45  30  54
1  30  55  29  60
2  28  29  33  51
3  36  56  37  62
4  29  40  27  73

# find min in every column
>>> df.min()
A    25
B    29
C    27
D    51
dtype: int64

# find max in every column
>>> df.max()
A    36
B    56
C    37
D    73
dtype: int64

# find min and max in every rows
>>> df.min(axis=1)
>>> df.max(axis=1)

# find max and min from whole dataframe
>>> np.min(df.min().to_numpy())
25
>>> np.max(df.max().to_numpy())
73

# find min and max in particular columns or rows
# max in column A
>>> df['A'].max()
36
# max in row 2
>>> df.loc[1].max()
60
# multiple columns
>>> df[['A', 'D']].max()
A    36
D    73
dtype: int64
# multiple rows (0 and 1 row)
>>> df.loc[0:1,].max(axis=1)
0    54
1    60
dtype: int64

# get row index of max and min values for each column
>>> df.idxmax()
A    3
B    3
C    3
D    4
dtype: int64
# for min df.idxmin()

# to skip missing values (NaN),  default skipna is True
# to not to skip missing values add skipna=False
>>> df.min(skipna=True)


Modifying dataframe columns and column names

import pandas as pd
df = pd.read_csv("https://reneshbedre.github.io/assets/posts/anova/onewayanova.txt", sep="\t")
df.head()
    A   B   C   D
0  25  45  30  54
1  30  55  29  60
2  28  29  33  51
3  36  56  37  62
4  29  40  27  73

# check column names
df.columns
Index(['A', 'B', 'C', 'D'], dtype='object')

# change column names
df.columns = ['A1', 'B1', 'C1', 'D1']
df.head()
    A1 	B1 	C1 	D1
0 	25 	45 	30 	54
1 	30 	55 	29 	60
2 	28 	29 	33 	51
3 	36 	56 	37 	62
4 	29 	40 	27 	73

# reorder column names
df1 = df[['B1', 'A1', 'C1', 'D1']]
# alternate way: df1 = df.reindex(columns=['B1', 'A1', 'C1', 'D1'])
df1
   B1  A1  C1  D1
0  45  25  30  54
1  55  30  29  60
2  29  28  33  51
3  56  36  37  62
4  40  29  27  73

# add new column
df['E']=[55,58,61,70,85]
df.head()
    A1 	B1 	C1 	D1 	E
0 	25 	45 	30 	54 	55
1 	30 	55 	29 	60 	58
2 	28 	29 	33 	51 	61
3 	36 	56 	37 	62 	70
4 	29 	40 	27 	73 	85

# drop column D1 and E
# axis=1 refers to column labels (use 0 for index labels)
df.drop(['D1', 'E'], axis=1)
    A1 	B1 	C1
0 	25 	45 	30
1 	30 	55 	29
2 	28 	29 	33
3 	36 	56 	37
4 	29 	40 	27

# rename specific columns
df.rename(columns={"A1": "a1", "B1": "b1"})
    a1 	b1 	C1 	D1 	E
0 	25 	45 	30 	54 	55
1 	30 	55 	29 	60 	58
2 	28 	29 	33 	51 	61
3 	36 	56 	37 	62 	70
4 	29 	40 	27 	73 	85

Correlation analysis with pandas dataframe

>>> import pandas as pd
>>> df = pd.read_csv("https://reneshbedre.github.io/assets/posts/anova/onewayanova.txt", sep="\t")
>>> df.head()
    A   B   C   D
0  25  45  30  54
1  30  55  29  60
2  28  29  33  51
3  36  56  37  62
4  29  40  27  73

# perform pearson correlation analysis
# check more methods at https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.corr.html
# returns correlation coefficient matrix
>>> df.corr(method='pearson')
          A         B         C         D
A  1.000000  0.569327  0.657544  0.341802
B  0.569327  1.000000  0.194615  0.264710
C  0.657544  0.194615  1.000000 -0.391609
D  0.341802  0.264710 -0.391609  1.000000

Compare all rows or columns

>>> import pandas as pd
>>> df = pd.DataFrame({'col1':['0', '0', '0' ], 'col2':[0,2,3], 'col3':[0,0.2,0.3]})
>>> df
  col1  col2  col3
0    0     0   0.0
1    0     2   0.2
2    0     3   0.3

# delete all rows where all columns are zero
>>> df[(df!=0).all(axis=1)]
  col1  col2  col3
1    0     2   0.2
2    0     3   0.3

# get all rows where columns sum > 3
>>> df[(df.sum(axis=1)>3)]
  col1  col2  col3
2    0     3   0.3

Enhance your skills with courses on Python and pandas

If you have any questions, comments or recommendations, please email me at reneshbe@gmail.com

This work is licensed under a Creative Commons Attribution 4.0 International License

Tags:

Updated: