23. Pandas: basic statistics

Let’s start by building a very sample dataframe.

import pandas as pd
df = pd.DataFrame()

names = ['Gandolf','Gimli','Frodo','Legolas','Bilbo']
types = ['Wizard','Dwarf','Hobbit','Elf','Hobbit']
magic = [10, 1, 4, 6, 4]
aggression = [7, 10, 2, 5, 1]
stealth = [8, 2, 5, 10, None]


df['names'] = names
df['type'] = types
df['magic_power'] = magic
df['aggression'] = aggression
df['stealth'] = stealth

Overview statistics

We can get an overview with the describe() method.

print (df.describe())

OUT:
       magic_power  aggression  stealth
count     5.000000    5.000000     4.00
mean      5.000000    5.200000     6.25
std       3.316625    3.420526     3.50
min       1.000000    2.000000     2.00
25%       4.000000    2.000000     4.25
50%       4.000000    5.000000     6.50
75%       6.000000    7.000000     8.50
max      10.000000   10.000000    10.00

We can modify the percentiles reported:

print (df.describe(percentiles=[0.05,0.1,0.9,0.95]))

OUT:

       magic_power  aggression  stealth
count     5.000000    5.000000     4.00
mean      5.000000    5.200000     6.25
std       3.316625    3.420526     3.50
min       1.000000    2.000000     2.00
5%        1.600000    2.000000     2.45
10%       2.200000    2.000000     2.90
50%       4.000000    5.000000     6.50
90%       8.400000    8.800000     9.40
95%       9.200000    9.400000     9.70
max      10.000000   10.000000    10.00

Specific statistics may be returned:

print (df.mean())

OUT:
magic_power    5.00
aggression     5.20
stealth        6.25
dtype: float64

List of key statistical methods

mean() = mean

median() = median

min() = minimum

max() =maximum

quantile(x)

var() = variance

std() = standard deviation

mad() = mean absolute variation

skew() = skewness of distribution

kurt() = kurtosis

cov() = covariance

corr() = Pearson Correlation coefficent

autocorr() = autocorelation

diff() = first discrete difference

cumsum() = cummulative sum

comprod() = cumulative product

cummin() = cumulative minimum

Returning the index of minimum and maximum

idxmin and idxmax will return the index row of the min/max. If two values are equal the first will be returned.

print ('Minimum:', df['aggression'].min())
print ('Index row:',df['aggression'].idxmin())
print ('\nFull row:\n', df.iloc[df['aggression'].idxmin()])

OUT:
Minimum: 2
Index row: 2

Full row:
 names           Frodo
type           Hobbit
magic_power         4
aggression          2
stealth             5
Name: 2, dtype: object

Removing rows with incomplete data

We can extract only those rows with a complete data set using the dropna() method.


print (df.dropna())

OUT:

     names    type  magic_power  aggression  stealth
0  Gandolf  Wizard           10           7      8.0
1    Gimli   Dwarf            1          10      2.0
2    Frodo  Hobbit            4           2      5.0
3  Legolas     Elf            6           5     10.0

We can use this directly in the describe method.

print (df.dropna().describe())

OUT:

    magic_power  aggression  stealth
count     4.000000    4.000000     4.00
mean      5.250000    6.000000     6.25
std       3.774917    3.366502     3.50
min       1.000000    2.000000     2.00
25%       3.250000    4.250000     4.25
50%       5.000000    6.000000     6.50
75%       7.000000    7.750000     8.50
max      10.000000   10.000000    10.00

To create a new dataframe with compplete rows only, we would simply assign to a new variable name:

df_na_dropped = df.dropna()

Counting number of different values in a column

print (df['type'].value_counts())

Out:

Hobbit    2
Wizard    1
Dwarf     1
Elf       1
Name: type, dtype: int64