Conditional lambda in pandas returns ValueError

In a df comprised of the columns asset_id, event_start_date, event_end_date, I wish to add a forth column datediff that for each asset_id will capture how many days passed between a end_date and the following start_date for the same asset_id, but in case that following start_date is earlier than the current end_date, I would like to capture the difference between the two start_dates. Dataset is sorted by (asset_id, start_date asc).

In Excel it would look something like:

I tried:

  events['datediff'] = df.groupby('asset_id').apply(lambda x:  x['event_start_date'].shift(-1)-x['event_end_date'] if
    x['event_start_date'].shift(-1)>x['event_end_date'] else  x['event_start_date'].shift(-1)-x['event_start_date'] ).
        fillna(pd.Timedelta(seconds=0)).reset_index(drop=True)

But this is:

1. not working. Throwing ValueError: The truth value of a Series is ambiguous.
2. so un-elegant.

Thanks!

Answer

df = pd.DataFrame({
    'asset_id':[0,0,1,1],
    'event_start_date':['2019-07-08','2019-07-11','2019-07-15','2019-07-25'],
    'event_end_date':['2019-07-08','2019-07-23','2019-07-29','2019-07-25']
                  })

df['event_end_date'] = pd.to_datetime(df['event_end_date'])
df['event_start_date'] = pd.to_datetime(df['event_start_date'])


df['next_start']=df.groupby('asset_id')['event_start_date'].shift(-1)

df['date_diff'] = np.where(
    df['next_start']>df['event_end_date'],
    (df['next_start']-df['event_end_date']).dt.days,
    (df['next_start']-df['event_start_date']).dt.days
        )

df = df.drop(columns=['next_start']).fillna(0)