[FIXED] Anova test in Python with a very large number of Groups

Issue

I have a relatively big dataset (approx 273,744 records) containing among others names of people and the dioptrics power they use:

Name   | Dioptric | Gender | Town |
-----------------------------------
'John' |  0.25    |   M    |  A   |
'Jack' |  0.5     |   M    |  C   |
'John' |  25      |   M    |  A   |
'Mary' |  0.25    |   F    |  C   |
........

I need to find if there is a correlation between name and dioptrics power. I decided to use the ANOVA test since there is one categorial and one quantitative variable. My problem is that the dataset contains a large number of name-dioptric groups (around 21,000) therefore I am not realy sure how to implement the

stats.f_oneway( Name_Dioptrics_GroupA, Name_Dioptrics_GroupB,....)

What I have done so far is:

• imported data as a numpy dataframe from the csv
• attempt to group based on name-dioptrics

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import scipy.stats as stats

# read data
data = pd.read_csv("dioptrics-to-name.csv")

# prepare data
dioptrics = data['value']
name = data['firstName']

"""
group based on name-dioptrics power
"""
name_dioptric_frame = pd.DataFrame({"Name":name,"dioptrics":dioptrics})
name_dioptrics_groups = name_dioptric_frame.groupby("Name").groups

## break into name-dioptrics groups
## name_dioptrics_GroupA = dioptrics[name_dioptrics_groups["John"]]
## name_dioptrics_GroupB = dioptrics[name_dioptrics_groups["Jamie"]] 
## and so on ....

print(stats.f_oneway( dioptrics[name_dioptrics_groups[ name_dioptrics_groups.keys()] ]) ) 
print(stats.f_oneway( dioptrics[name_dioptrics_groups[ [ name for x in name_dioptrics_groups() ] ] ]) ) 

It doesn't work of course... Am I taking a correct approach here?

Solution

Pandas groupby function allows you to group your dataframe by several columns. You can use this feature if you use a list of columns instead of one column:

df = pd.DataFrame([
    ['WAKA', 2, '1'],
    ['WAKA-WAKA', 3, '7'],
    ['WAKKA', 1, '0'],
    ['WAKA', 2, '1'],
    ['WAKA-WAKA', 1, '7'],
    ['WAKKA', 1, '1'],
    ['WAKA', 5, '1'],
    ['WAKA-WAKA', 3, '7'],
    ['WAKKA', 1, '2'],
])
df.columns = ['name', 'd', 'info']

df.groupby(['name', 'd']).groups

Will return:

{('WAKA', 2): Int64Index([0, 3], dtype='int64'),
 ('WAKA', 5): Int64Index([6], dtype='int64'),
 ('WAKA-WAKA', 1): Int64Index([4], dtype='int64'),
 ('WAKA-WAKA', 3): Int64Index([1, 7], dtype='int64'),
 ('WAKKA', 1): Int64Index([2, 5, 8], dtype='int64')}

In your code you are trying to group by only name, without dioptrics.

Answered By - vurmux