Assume I have a 1D array and want to replace / interpolate NaN blocks of length n with copies of the n/2 non-nan previous values and the n/2 non-nan subsequent values.
Example 1:
input = [1, 2, NaN, NaN, NaN, NaN, 3, 2]
output= [1, 2, 1, 2, 3, 2, 3, 2]
Example 2: if n is odd, fill with n%2+1 previous values and n%2 subsequent values
input = [1, 2, 3, NaN, NaN, NaN, 4, 2]
output= [1, 2, 3, 2, 3, 4, 4, 2]
Example 3: if not enough non-nan neighbours are available, replicate the available one (in this example value = 3)
input = [3, NaN, NaN, NaN, NaN, 4, 2]
output = [3, 3, 3, 4, 2, 4, 2]
my prelim solution looks like this..
def fillna_with_neighbours(data):
# get start / stop of nan blocks
nan_blocks = np.where(np.isnan(data),1,0)
nan_blocks = np.concatenate([[0],nan_blocks,[0]])
nan_blocks = np.where(np.diff(nan_blocks)!=0)[0]
nan_blocks = nan_blocks.reshape(-1, 2)
for block in nan_blocks:
nan_start, nan_end = block
n = nan_end - nan_start
n_pre = n//2 if n%2 == 0 else n//2 + 1
nan_pre = nan_start - n_pre if nan_start>=n_pre else 0
n_post = n//2
nan_post = nan_end + n_post
pre = data[nan_pre:nan_start]
post = data[nan_end:nan_post]
if pre.size < n_pre:
pre = np.resize(pre, n_pre)
if post.size < n_post:
post = np.resize(post, n_post)
data[nan_start:nan_end] = np.hstack([pre, post])
return data
ex1 = np.asarray([1, 2, np.nan, np.nan, np.nan, np.nan, 3, 2])
ex2 = np.asarray([1, 2, 3, np.nan, np.nan, np.nan, 4, 2])
ex3 = np.asarray([3, np.nan, np.nan, np.nan, np.nan, 4, 2])
is there any ready (scipy?) function for this problem. I am sure there are much better
