Gets 2 maximum element form max pooling

Pytorch_get2max.py

import numpy as np
y = np.array([[1.0, 7.0, 1.0],[3.0, 4.0, 2], [1, 7, 1]])
x = torch.from_numpy(y)
x = Variable(x)
x = x.resize(1, 3, 3)
stride = 2
new_var = Variable(torch.zeros([x.shape[0], x.shape[1]//stride, x.shape[2]//stride]))
new_var2 = Variable(torch.zeros([x.shape[0], x.shape[1]//stride, x.shape[2]//stride]))
for dim1 in range(x.shape[0]):
    tmp = Variable(torch.zeros([x.shape[1]//stride, x.shape[2]//stride, 1]))
    tmp2 = Variable(torch.zeros([x.shape[1]//stride, x.shape[2]//stride, 1]))
    for i in range(0, x.shape[1], stride):
        for j in range(0, x.shape[2], stride):
            tmp_max = x[dim1][i][j]
            tmp_max2 = x[dim1][i][j]
            for k in range(stride):
                for m in range(stride):
                    if (i+k < x.shape[1] and j+m < x.shape[2]):
                        tmp_max = torch.max(tmp_max, x[dim1][i+k][j+m])
            for k in range(stride):
                for m in range(stride):
                    if (i+k < x.shape[1] and j+m < x.shape[2]):
                        if (tmp_max.data > x[dim1][i+k][j+m].data).all():
                            if (tmp_max2.data < x[dim1][i+k][j+m].data).all():
                                tmp_max2 = x[dim1][i+k][j+m]
            if (i//stride < tmp.shape[0] and j//stride < tmp.shape[1]):
                tmp[i//stride, j//stride] = tmp_max
                tmp2[i//stride, j//stride] = tmp_max2
    new_var[dim1] = tmp
    new_var2[dim1] = tmp2