def linearInterpolation(db):
for i in range(0, db.shape[1]):
ts = db[:,i]
seq = sq.detectSequences(ts)
if len(seq)>1: ##In case there are more than one sequence
seq = sq.joinSequences(seq) ##Join sequences in case they are relatively close
seq = pointSelector(seq) ##Select points
for j in range(0, len(seq)):
s = pd.Series(ts[seq[j][0]:seq[j][1]])
db[seq[j][0]:seq[j][1],i] = (s.interpolate(method = "linear")).values
##In case values are lost at the beggining or the end of the TS,
db = (pd.DataFrame(db)).fillna(method='pad').as_matrix()
db = (pd.DataFrame(db)).fillna(method='bfill').as_matrix()
return(db)
def multInterpolation(db):
for i in range(0, db.shape[1]):
ts = db[:,i]
seq = sq.detectSequences(ts)
if len(seq)>1: ##In case there are more than one sequence
seq = sq.joinSequences(seq) ##Join sequences in case they are relatively close
seq = pointSelector(seq) ##Select points
for j in range(0, len(seq)):
s = pd.Series(ts[seq[j][0]:seq[j][1]])
##Three types are considered, combining to a simple Multiple Imputation Algorithm.
s1 = s.interpolate(method = "polynomial", order = int(min(s.notnull().sum()-1, 4)))
s2 = s.interpolate(method = "spline", order = int(min(s.notnull().sum()-1, 3)))
s3 = s.interpolate(method = "cubic")
db[seq[j][0]:seq[j][1],i] = (s1.values + s2.values + s3.values)/3
##In case values are lost at the beggining or the end of the TS,
db = (pd.DataFrame(db)).fillna(method='pad').as_matrix()
db = (pd.DataFrame(db)).fillna(method='bfill').as_matrix()
return(db)