我尝试使用岭校正来执行多元线性回归，以确定xarray Dataframe中某些空间变量之间的关系。因为这些是观测数据，所以数据中偶尔会有NaN值，sklearn无法原生处理。我尝试使用data.interpolate_na(fill_value='extrapolate')，但这无法替换所有NaN值。一个可行的解决方案是使用data.fillna(0)，但这可能会有问题，因为我宁愿不'发明'数据从整个布时，像插值或掩蔽会更好。我的代码如下：

from sklearn.linear_model import Ridge
import xarray as xr
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import cartopy.crs as ccrs

def detrend_dim(da, dim, deg=1): # removes (temporal) trends from data
    # detrend along a single dimension
    p = da.polyfit(dim=dim, deg=deg)
    fit = xr.polyval(da[dim], p.polyfit_coefficients)
    return da - fit

#loading and preprocessing data
data = xr.open_dataset('reanalysisdata.nc')
data2 = xr.open_dataset('satellitedata.nc')
land = xr.open_dataset('twoptfivedeglandmask.nc')
data2.coords['mask'] = (('lat', 'lon'), land.FRLAND.mean(dim='time').data)
data2 = data2.where(data2.mask == 0)
data.coords['mask'] = (('lat', 'lon'), land.FRLAND.mean(dim='time').data)
data = data.where(data.mask == 0)

#trying to deal with NaN values
data = data.fillna(0)
data2 = data2.fillna(0)

#removing time trends
data['x1'] = detrend_dim(data.x1, 'time')
data['x2'] = detrend_dim(data.x2, 'time')
data['x3'] = detrend_dim(data.x3, 'time')
data2['y'] = detrend_dim(data2.y, 'time')

#removing seasonal cycle
dataM = data.groupby('time.month')
dataU = dataM - dataM.mean()
data2M = data2.groupby('time.month')
data2U = data2M - data2M.mean()

#performing regression
a = np.zeros((72,144,3))
for i in range(len(data.lat)):
    for j in range(len(data.lon)):
        a[i,j,:] = (Ridge().fit(np.array((dataU.isel(lev=2).x1.values[:,i,j].reshape(-1,1),
                    dataU.isel(lev=2).x2.values[:,i,j].reshape(-1,1),
                    dataU.isel(lev=2).x3.values[:,i,j].reshape(-1,1))).reshape(108,3),
                    data2U.y.values[:,i,j].reshape(108)).coef_)
dataU = data.assign_coords(varname=['x1','x2','x3'])
dataU['multiple_reg_coeff'] = (('lat','lon','varname'), a)

字符串
我需要迭代地检查每个变量的NaN吗？在sklearn.linear_model中有回归，它可以原生地处理NaN，但我对它们背后的数学知识的理解比岭回归少。