Python :Nested for loops fail on the second loop

Question

I am trying to divide a large data set into smaller parts for an analysis. I been using a for-loop to divide the data set before implementing the decision trees. Please see a small version of the data set below:

ANZSCO4_CODE          Skill_name              Cluster         date
  1110                  computer                 S              1
  1110                  communication            C              1
  1110                  SAS                      S              2
  1312                  IT support               S              1
  1312                  SAS                      C              2
  1312                  IT support               S              1
  1312                  SAS                      C              1

First step I create an empty dictionary:

d = {}

and the lists:

 list = [1110, 1322, 2111]
 s_type = ['S','C']

Then run the following loop:

for i in list:
    d[i]=pd.DataFrame(df1[df1['ANZSCO4_CODE'].isin([i])] )

The result is a dictionary with 2 data sets inside.

As a next step I would like to subdivide the data sets into S and C. I run the following code:

for i in list:
    d[i]=pd.DataFrame(df1[df1['ANZSCO4_CODE'].isin([i])] )

    for b in s_type:
         d[i]=  d[i][d[i]['SKILL_CLUSTER_TYPE']==b]

As a final result I would expect to have 4 separate data sets, being: 1110 x S, 1110 x C , 1312 x S and 1312 and C.

However when I implement the second code I get only 2 data sets inside the dictionary and they are empty.

Answer 1

I think there was empty DataFrames, because in data was not values from list called L (Dont use variable name list, because python reserved word).

from  itertools import product

L = [1110, 1312, 2111]
s_type = ['S','C']

Then create all combinations all lists:

comb = list(product(L, s_type))
print (comb)
[(1110, 'S'), (1110, 'C'), (1312, 'S'), (1312, 'C'), (2111, 'S'), (2111, 'C')]

And last create dictionary of DataFrames:

d = {}
for i, j in comb:
    d['{}x{}'.format(i, j)] = df1[(df1['ANZSCO4_CODE'] == i) & (df1['Cluster'] == j)]

Or use dictionary comprehension:

d = {'{}x{}'.format(i, j): df1[(df1['ANZSCO4_CODE'] == i) & (df1['Cluster'] == j)] 
      for i, j in comb}

---

print (d['1110xS'])
   ANZSCO4_CODE Skill_name Cluster
0          1110   computer       S
2          1110        SAS       S

EDIT:

If need all combinations of possible data by columns use groupby:

d = {'{}x{}x{}'.format(i,j,k): df2 
      for (i,j, k), df2 in df1.groupby(['ANZSCO4_CODE','Cluster','date'])}
print (d)
{'1110xCx1':    ANZSCO4_CODE     Skill_name Cluster  date
1          1110  communication       C     1, '1110xSx1':    ANZSCO4_CODE Skill_name Cluster  date
0          1110   computer       S     1, '1110xSx2':    ANZSCO4_CODE Skill_name Cluster  date
2          1110        SAS       S     2, '1312xCx1':    ANZSCO4_CODE Skill_name Cluster  date
6          1312        SAS       C     1, '1312xCx2':    ANZSCO4_CODE Skill_name Cluster  date
4          1312        SAS       C     2, '1312xSx1':    ANZSCO4_CODE  Skill_name Cluster  date
3          1312  IT support       S     1
5          1312  IT support       S     1}

print (d.keys())
dict_keys(['1110xCx1', '1110xSx1', '1110xSx2', '1312xCx1', '1312xCx2', '1312xSx1'])

Another different approach is if need processes each group is use GroupBy.apply:

def func(x):
    print (x)
    #some code for process each group
    return x

   ANZSCO4_CODE     Skill_name Cluster  date
1          1110  communication       C     1
   ANZSCO4_CODE     Skill_name Cluster  date
1          1110  communication       C     1
   ANZSCO4_CODE Skill_name Cluster  date
0          1110   computer       S     1
   ANZSCO4_CODE Skill_name Cluster  date
2          1110        SAS       S     2
   ANZSCO4_CODE Skill_name Cluster  date
6          1312        SAS       C     1
   ANZSCO4_CODE Skill_name Cluster  date
4          1312        SAS       C     2
   ANZSCO4_CODE  Skill_name Cluster  date
3          1312  IT support       S     1
5          1312  IT support       S     1

df2 = df1.groupby(['ANZSCO4_CODE','Cluster','date']).apply(func)
print (df2)