python - Pandas self-conditional column -

i trying create column in pandas conditional it's previous value along other columns.

import pandas pd import numpy np = np.random.standard_normal(100) = pd.dataframe(a) a['out'] = 0 a['out2'] = 0 t in range(1,a.shape[0]):     if (a[0][t] > 1) & (a['out'][t-1]==0):         a['out'][t] = 1     elif (a[0][t] < -1) & (a['out'][t-1]==0):         a['out'][t] = -1     elif ((a[0][t] > 0) & (a['out'][t-1]==-1)) | ((a[0][t] < 0) & (a['out'][t-1]==1)):         a['out'][t] = 0     else:         a['out'][t] = a['out'][t-1]  a['out2'] = np.where((a.index== 0),0         ,np.where((a[0] > 1) & (a['out2'].shift()==0), 1         ,np.where((a[0] < -1) & (a['out2'].shift()==0), -1         ,np.where(((a[0] > 0) & (a['out2'].shift()==-1)) | ((a[0] < 0) & (a['out2'].shift()==1)), 0         ,a['out2'].shift())))) 

the column a['out2'] tries copy a['out'] in vectorized form not read previous values. column a['out'] takes long compile through loop. can me faster, vectorized approach creating column?

you can create function , use apply. access previous data can use variable store value. hope following code helps.

import pandas pd import numpy np = np.random.standard_normal(100) = pd.dataframe(a) state = 0 def get_val(a,prev_state):     global state     if (a > 1) & (prev_state==0):         state = 1     elif (a < -1) & (prev_state==0):         state = -1     elif ((a > 0) & (prev_state==-1)) | ((a < 0) & (prev_state==1)):         state = 0          return state  a['out'] = a[0].apply(lambda x: get_val(x,state)) 

output :

            0  out  0  1.366864    1      1  0.887763    1      2 -0.663636    0      3 -1.824950   -1      4  0.459663    0     5 -1.325129   -1      6  1.587188    0     7 -0.148159    0      8  0.578862    0      9  0.758460    0      

if use %%timeit

100 loops, best of 3: 2.16 ms per loop