fill_values.py

from scipy import interpolate

def fill_values(data, 
                max_contiguous_na=5, 
                periods_for_extrapolation=5, 
                method='slinear'):
    time = np.arange(len(data))
    signal = data.copy()
    # If there are no valid data points, there's nothing to fill
    if np.all(np.isnan(signal)):
        return signal
    elif np.any(np.isnan(signal)):
        # Gets minimum and maximum values
        minv, maxv = np.nanmin(signal), np.nanmax(signal)
        # Uses only the values in the longest sequence containing no more than
        # a given number of contiguous missing points (5)
        ini, end = bounds(data, max_contiguous_na)
        signal = signal[ini:end]
        
        # Creates a filler function to interpolate the missing values over time
        try:
            filler_func = interpolate.interp1d(time[ini:end][~np.isnan(signal)], 
                                               signal[~np.isnan(signal)], 
                                               kind=method)
        except ValueError:
            return signal
        
        # Passes the time as argument to the filler function
        filled = filler_func(time[ini:end])
        # Caps interpolated values at actually observed min and max values
        filled = np.minimum(filled, maxv)
        filled = np.maximum(filled, minv)
        if ini > 0:
            # Rebuilds the full sized array, if the longest sequence 
            # doesn't start at zero
            filled = np.concatenate([[np.nan] * ini, filled])
        if end < len(data):
            # If the longest sequence ends before the last time period, 
            # extrapolates the last missing values using the average of
            # a given number of periods
            avg = filled[-periods_for_extrapolation:].mean()
            filled = np.concatenate([filled, 
                                     np.nan_to_num(data[end:].copy(), nan=avg)])
        return filled
    else:
        return signal