插入具有随机缺失列的矩阵

Question

我想插入一个随机缺失列的矩阵。

我创建了一个包含随机条目（可以是图像）的矩阵，将一些列置零，然后它们尝试对置零的列进行插值。

import numpy as np 
from scipy.interpolate import griddata

l = 2000
c = 340

tdgn = np.random.rand(l, c) #array with 2000 lines and 240 columns, or some grayscale image, binary file... np.load(...)


#sort some column indexes to "decimate"
maxindex=tdgn.shape[1]
p_dec = 0.3
index = np.arange(maxindex)
index_dec = np.sort(np.random.choice(maxindex, int(np.round(maxindex*p_dec)), replace=False))
index_nondec = np.array(list(filter(lambda x: x not in index_dec, index)))

#decimate
tdgn_dec = np.copy(tdgn)
tdgn_dec[:, index_nondec] = 0

#grid points to interpolate
grid_coods_dec_x, grid_coods_dec_y = np.meshgrid(np.arange(tdgn_dec.shape[0]), index_dec )
grid_coods_x, grid_coods_y = np.meshgrid(np.arange(tdgn.shape[0]), np.arange(tdgn.shape[1]) )


#interpolate
tdgn_interp = griddata( (grid_coods_dec_x, grid_coods_dec_y), tdgn[ grid_coods_dec_x, grid_coods_dec_y], (grid_coods_x, grid_coods_y), method='linear')

'tdgn_interp' 应与 'tdgn_dec' 具有相同的维度，但其归零值应尽可能接近原始矩阵 'tdgn'。

但是，"ValueError: invalid shape for input data points" 返回...

Answer 1

def interp_traces(tdgn_dec, col_nondec, method='linear'):

    grid_coods_x, grid_coods_y = np.meshgrid(np.arange(tdgn_dec.shape[1]), np.arange(tdgn_dec.shape[0]) )

    grid_coods_nondec_x, grid_coods_nondec_y = np.meshgrid(col_nondec, np.arange(tdgn_dec.shape[0]) )

    f_grid_x = grid_coods_nondec_x.flatten()
    f_grid_y = grid_coods_nondec_y.flatten()

    f_points = tdgn_dec[grid_coods_nondec_y, grid_coods_nondec_x].flatten()

    interp = griddata((f_grid_x, f_grid_y), f_points, (grid_coods_x, grid_coods_y), method=method)

    return( interp )

如上所述使用 'interp_traces'...

tdgn_interp = interp_traces(tdgn_dec, index_nondec)

插入具有随机缺失列的矩阵

Interpolate a matrix with random missing columns

interpolation

numpy

scipy

python-3.x