python - split into training set and test set with specific attribute values for rows

Question

My input file is under the following form:

gold,Attribute1,Attribute2
T,1,1
T,1,2
T,1,1
N,1,2
N,2,1
T,2,1
T,2,2
N,2,2
T,3,1
N,3,2
N,3,1
T,3,2
N,3,3
N,3,3

I am trying to predict the first column using the second and third columns. I would like to split this input data randomly into a training set and a test set such that all the rows having a specific combination of the values of <attribute1, attribute2> fall either in the test set or the training set. For example, all the rows with values <1,1>, <1,2>, <2,1> should fall into the training set and all the rows with values <2,2>, <3,1>, <3,2>, <3,3> should fall in the test set. This has to be made randomly, this was just an example. How can I make such a split?

question from:https://stackoverflow.com/questions/65933830/split-into-training-set-and-test-set-with-specific-attribute-values-for-rows

Answer 1

A simple way of spliting this will be through condition rather than pre-defined methods.

Code :-

import numpy as np
import pandas as pd 

df = pd.DataFrame(pd.read_csv('test.csv'))

print(df.head())
print(df.describe())
print(type(df['Attribute1']))

#For only getting values where both are less than 2 or equal to 2
df_Condition1 = df[df['Attribute1'] <= 2]
Train_Set = df_Condition1[df_Condition1['Attribute2'] <= 2]

#to subract the remaining elements 
Test_Set = df[ df.isin(Train_Set) == False]
Test_Set =Test_Set.dropna()

print(Train_Set)
print(Test_Set)

Output :

   gold  Attribute1  Attribute2
   0    T           1           1
   1    T           1           2
   2    T           1           1
   3    N           1           2
   4    N           2           1
  
   Attribute1  Attribute2
   count   14.000000   14.000000
   mean     2.142857    1.714286  
   std      0.864438    0.726273
   min      1.000000    1.000000 
   25%      1.250000    1.000000
   50%      2.000000    2.000000
   75%      3.000000    2.000000
   max      3.000000    3.000000
   <class 'pandas.core.series.Series'>

       gold  Attribute1  Attribute2
   0    T           1           1
   1    T           1           2
   2    T           1           1
   3    N           1           2
   4    N           2           1
   5    T           2           1
   6    T           2           2
   7    N           2           2

      gold  Attribute1  Attribute2
   8     T         3.0         1.0
   9     N         3.0         2.0
   10    N         3.0         1.0
   11    T         3.0         2.0
   12    N         3.0         3.0
   13    N         3.0         3.0