[FEA] Linear interpolation of missing values

[beckernick]

In time series analyses, I may have missing values throughout my data that I would like to fill. In the general case, I can use Series.fillna to fill missing values with things like a scalar, the preceding valid value (forward fill), or the next valid value (backward fill).

When my data has a temporal trend, the effectiveness of these techniques can break down. In such situations, I often want to interpolate between the valid values to replace my missing values with a statistical approach rather than fill missing values with a single scalar, forward, or backward fill.

Pandas provides such functionality via an interpolate API that delegates to numpy.interp and the scipy.interpolate sub-package to support a variety of interpolation techniques (including the standard forward fill described above).

I'd like to be able to able to conduct linear interpolation like I can do in pandas. For linear interpolation, pandas delegates to numpy.interp which has an corresponding implementation in cupy (meaning this may be possible in the Python layer)

import pandas as pd
​
s = pd.Series([0, 2, None, None, None, 8])
s.interpolate(method='linear')
0    0.0
1    2.0
2    3.5
3    5.0
4    6.5
5    8.0
dtype: float64