Condition Shift in Pandas

I am trying to count up a number during a sequence change.

The number shall always be +1, when changing from the negative to the positive range.

Here the code:

data = {'a':  [-1,-1,-2,-3,4,5,6,-2,-2,-3,6,3,6,7,-1,-5,-7,1,34,5]}

df = pd.DataFrame (data)


df['p'] = df.a > 0
df['g'] = (df['p'] != df['p'].shift()).cumsum()

This is the output:

0  -1  False  1
1  -1  False  1
2  -2  False  1
3  -3  False  1
4   4   True  2
5   5   True  2
6   6   True  2
7  -2  False  3
8  -2  False  3
9  -3  False  3
10  6   True  4
11  3   True  4
12  6   True  4
13  7   True  4
14 -1  False  5

I need an output that looks like this:

0  -1  False  1
1  -1  False  1
2  -2  False  1
3  -3  False  1
4   4   True  2
5   5   True  2
6   6   True  2
7  -2  False  2
8  -2  False  2
9  -3  False  2
10  6   True  3
11  3   True  3
12  6   True  3
13  7   True  3
14 -1  False  3

Anybody got an idea?