scikit learn - Missing value imputation in python using KNN -

i have dataset looks this

1908    january 5.0 -1.4 1908    february    7.3 1.9 1908    march   6.2 0.3 1908    april   nan   2.1 1908    may nan   7.7 1908    june    17.7    8.7 1908    july    nan   11.0 1908    august  17.5    9.7 1908    september   16.3    8.4 1908    october 14.6    8.0 1908    november    9.6 3.4 1908    december    5.8 nan 1909    january 5.0 0.1 1909    february    5.5 -0.3 1909    march   5.6 -0.3 1909    april   12.2    3.3 1909    may 14.7    4.8 1909    june    15.0    7.5 1909    july    17.3    10.8 1909    august  18.8    10.7   

i want replace nans using knn method. looked sklearns imputer class supports mean, median , mode imputation. there feature request here don't think thats been implemented of now. ideas on how replace nans last 2 columns using knn?

edit: since need run codes on environment, don't have luxury of installing packages. sklearn, pandas, numpy , other standard packages ones can use.

fancyimpute package supports such kind of imputation, using following api:

from fancyimpute import knn     # x complete data matrix # x_incomplete has same values x except subset have been replace nan  # use 3 nearest rows have feature fill in each row's missing features x_filled_knn = knn(k=3).complete(x_incomplete) 

here imputations supported package:

•simplefill: replaces missing entries mean or median of each column.

•knn: nearest neighbor imputations weights samples using mean squared difference on features 2 rows both have observed data.

•softimpute: matrix completion iterative soft thresholding of svd decompositions. inspired softimpute package r, based on spectral regularization algorithms learning large incomplete matrices mazumder et. al.

•iterativesvd: matrix completion iterative low-rank svd decomposition. should similar svdimpute missing value estimation methods dna microarrays troyanskaya et. al.

•mice: reimplementation of multiple imputation chained equations.

•matrixfactorization: direct factorization of incomplete matrix low-rank u , v, l1 sparsity penalty on elements of u , l2 penalty on elements of v. solved gradient descent.

•nuclearnormminimization: simple implementation of exact matrix completion via convex optimization emmanuel candes , benjamin recht using cvxpy. slow large matrices.

•biscaler: iterative estimation of row/column means , standard deviations doubly normalized matrix. not guaranteed converge works in practice. taken matrix completion , low-rank svd via fast alternating least squares.