如何处理具有张量流的嵌套循环？

i am new to tensorflow. I am working with keras but for creating a customized loss function i am more or less forced to write a function in tensorflow. I get stuck at the point where i have to translate this following numpy for loop into tensorflow syntax.

我是tensorflow的新手。我正在使用keras但是为了创建一个自定义的损失函数,我或多或少*在tensorflow中编写一个函数。我陷入困境,我必须将numpy for循环转换为tensorflow语法。

for j in range(grid):
    for k in range(modes):
        for l in range(dim):
            for m in range(dim):
                lorentz[:,j,l,m] += 1J*osc_stre[:,l,m,k]/(energies[j]-e_j[:,k])
                if l == m == k:
                    lorentz[:,j,l,m] += 1

Here you can see the initial shapes of the arrays:

在这里,您可以看到数组的初始形状:

e_j = zeros([sample_nr,modes],dtype='complex')
osc_stre = zeros([sample_nr,dim,dim,modes],dtype='complex')
lorentz = zeros([sample_nr,grid,dim,dim],dtype='compex')

energies[j] has the shape (grid)

能量[j]有形状(网格)

is it possible to handle this problem with tensorflow? Can anybody give me a hint how to translate this into tensorflow syntax? I have already tried a couple of things like the tensorflow while loop but one of the big problems is, that tensorflow objects do not support item assignment.

是否有可能用tensorflow处理这个问题?任何人都可以给我一个如何将其转换为tensorflow语法的提示吗?我已经尝试了一些像tensorflow while循环这样的东西,但其中一个大问题是,tensorflow对象不支持项目赋值。

EDIT:

i think i've come up with a solution for this simplified verison of the problem:

我想我已经为这个简化的问题验证提出了一个解决方案:

for j in range(grid):
    for k in range(modes):
        lorentz[j] += 1J*osc_stre[k]/(energies[j]-e_j[k])
        if k == 0:
           lorentz[j] += 1

the solution:

lorentz_list = []
    tf_one = tf.ones([1], complex64)
    tf_i = tf.cast(tf.complex(0.,1.), complex64)
    energies_float = tf.cast(energies,float32)
    energies_complex = tf.complex(energies_float,tf.zeros([energy_grid],float32))
    for j in range(energy_grid):
        lorentz_list.append(tf.add(tf_one,tf.reduce_sum(tf.multiply(tf_i,tf.divide(osc_stre_tot,tf.subtract(energies_complex[j],e_j))),-1)))
    lorentz = tf.stack(lorentz_list)

1 个解决方案

#1

Assuming these:

lorentz.shape == (batch, grid, dim, dim) and was zero before the loop.

lorentz.shape ==(batch,grid,dim,dim)并且在循环之前为零。

osc_stre.shape == (batch, dim, dim, modes)

osc_stre.shape ==(批量,暗淡,昏暗,模式)

energies.shape == (grid,)

energies.shape ==(网格,)

e_j.shape == (batch, modes)

e_j.shape ==(批量,模式)

Then:

osc_stre = K.reshape(osc_stre, (-1, 1, dim, dim, modes))
energies = K.reshape(energies, (1, grid, 1, 1, 1))   
e_j = K.reshape(e_j, (-1, 1, 1, 1, modes))

lorentz = 1J*osc_stre/(energies-e_j) 

identity = np.zeros((1, 1, dim, dim, modes))
for d in range(min(modes,dim)):
    identity[0,0,d,d,d] = 1
identity = K.variable(identity, dtype = tf.complex64)

lorentz += identity
lorentz = K.sum(lorentz, axis=-1)

#1