How can I use transform (or other) instead of merge and temporary columns for special indexing/ranking?

I have the following python code which outputs correctly, but I sense the code could be more pythonic by removing temporary columns and removing the merge perhaps with transform or other clever way. The data is presorted by category then subcategory. Each group will have a will a sub_cat value of 'c'. The idea is to output a rank column where sub_cat value 'c' is always zero and counts by integers up/down through the rest of the group, resetting at each group. The cat_index, and c_idx (meaning the index of 'c' within the group) will be disposed of.

Here is my working code:

import pandas as pd
import io

data = '''
category,sub_cat
A,a
A,c
A,d
B,c
B,d
D,a
D,b
D,c
G,b
G,c
G,e
G,f
G,h
'''

df = pd.read_csv(io.StringIO(data), sep=',')
df['cat_index'] = df.groupby('category').cumcount()
c_idx = df.groupby('category').apply(lambda x: x.loc[x['sub_cat'] == 'c', 'cat_index'].values[0]).to_frame('c_idx')
df = df.merge(c_idx, how='left', on='category')
df['rank'] = df['cat_index'] - df['c_idx']

And here is the correct output:

Again, 'rank' is the desired output from inputs 'category', and 'sub_cat'. Columns 'cat_index' and 'c_idx' are temporary and will be dropped.

Any ideas? Let me know. Thanks.

Try:

df['rank'] = df.groupby('category')['sub_cat'].transform(lambda x: np.arange(len(x)) - ((x == 'c').idxmax() - x.index[0]))
print(df)

Prints:

   category sub_cat  rank
0         A       a    -1
1         A       c     0
2         A       d     1
3         B       c     0
4         B       d     1
5         D       a    -2
6         D       b    -1
7         D       c     0
8         G       b    -1
9         G       c     0
10        G       e     1
11        G       f     2
12        G       h     3