pre_processing.py 文件源码

def interpolate_missing_data(Y):
    """
    Interpolate any missing data using nearest neighbor interpolation.
    Missing data is identified as entries with values NaN
    Input:
    Y   np.ndarray (3D)
        movie, raw data in 3D format (d1 x d2 x T)

    Outputs:
    Y   np.ndarray (3D)
        movie, data with interpolated entries (d1 x d2 x T)
    coor list
        list of interpolated coordinates
    """
    coor=[];
    if np.any(np.isnan(Y)):
        raise Exception('The algorithm has not been tested with missing values (NaNs). Remove NaNs and rerun the algorithm.')
        # need to
        for idx,row in enumerate(Y):
            nans=np.where(np.isnan(row))[0]
            n_nans=np.where(~np.isnan(row))[0]
            coor.append((idx,nans))
            Y[idx,nans]=np.interp(nans, n_nans, row[n_nans])


#    mis_data = np.isnan(Y)
#    coor = mis_data.nonzero()
#    ok_data = ~mis_data
#    coor_ok = ok_data.nonzero()
#    Yvals=[np.where(np.isnan(Y)) for row in Y]
#
#    Yvals = griddata(coor_ok,Y[coor_ok],coor,method='nearest')
#    un_t = np.unique(coor[-1])
#    coorx = []
#    coory = []
#    Yvals = []
#    for i, unv in enumerate(un_t):
#        tm = np.where(coor[-1]==unv)
#        coorx.append(coor[0][tm].tolist())
#        coory.append(coor[1][tm].tolist())
#        Yt = Y[:,:,unv]
#        ok = ~np.isnan(Yt)
#        coor_ok = ok.nonzero()
#        ytemp = griddata(coor_ok,Yt[coor_ok],(coor[0][tm],coor[1][tm]),method='nearest')
#        Yvals.append(ytemp)

    return Y, coor

#%%