Create list from value in current row and previous row based on condition

Question

I have a dataframe with two columns 'a' and 'b' where 'b' is the difference between the value of 'a' and the previous value 'a'

df = pd.DataFrame({'a': [10, 60, 30, 80, 10]})
df['b'] = df['a']-df['a'].shift(1)

    a   b
0   10  NaN
1   60  50.0
2   30  -30.0
3   80  50.0
4   10  -70.0

I want to create a new column 'c' with values as a list of previous value of 'a' and the current value of 'a' (example, [60,30]) only where the column 'b' is negative. Otherwise it has to be a list of the current value 'a' itself.

The resulting output should look like

    a   b       c
0   10  NaN     [10]
1   60  50.0    [60]
2   30  -30.0   [60, 30]
3   80  50.0    [80]
4   10  -70.0   [80, 10]

Answer 1

Use list comprehension for create lists if b < 0 in numpy array with shifted helper column s by Series.shift added by DataFrame.assign :

arr = df.assign(s = df['a'].shift(fill_value=0))[['a','b','s']].to_numpy()
df['c'] = [[s,a] if b < 0 else [a] for  a,b,s in arr]
print (df)
    a     b             c
0  10   NaN        [10.0]
1  60  50.0        [60.0]
2  30 -30.0  [60.0, 30.0]
3  80  50.0        [80.0]
4  10 -70.0  [80.0, 10.0]  

Or is used Series.mask with one element list created by list comprenension:

s = pd.Series([[x] for x in df['a']], index=df.index)
#alternative
s = df['a'].apply(lambda x: [x])

---

df['c'] = s.mask(df['b'].lt(0), s.shift() + s)
print (df)
    a     b         c
0  10   NaN      [10]
1  60  50.0      [60]
2  30 -30.0  [60, 30]
3  80  50.0      [80]
4  10 -70.0  [80, 10]

Answer 2

Use Series.to_numpy and increase the dimension by adding the newaxis then use boolean indexing with Series.lt and assign the new values:

df['c'] = df['a'].to_numpy()[:, None].tolist()
df.loc[df['b'].lt(0), 'c'] = df['c'].shift() + df['c']

Result:

   a     b         c
0  10   NaN      [10]
1  60  50.0      [60]
2  30 -30.0  [60, 30]
3  80  50.0      [80]
4  10 -70.0  [80, 10]

Answer 3

Load the data:

df = pd.DataFrame({'a': [10, 60, 30, 80, 10]})
df['b'] = df['a']-df['a'].shift(1)

Create a temporary Numpy matrix:

npa = np.array([df['a'].shift(1), df['a']]).transpose()

Write the matrix to a new df column 'c':

df['c'] = list(npa)

Copy values in 'a' to 'c' if values in column 'b' are larger than 0 or NAN:

df.loc[(df['b'] > 0) | (df['b'].isnull() == True) , 'c'] = pd.Series([[x] for x in df['a']])