What is en_ksize/de_ksize and max_diffusion_step?

TensorFlow General Discussion
1

Hi!

I am learning how to change parameters of a model to train. However, I do not know what en_ksize/de_ksize and max_diffusion_step are, not even sure what they are for. Thanks

en_ksize: [3, 3, 3]
de_ksize: [3, 3, 3]
max_diffusion_step: 1

3

@Lara,

Could you please share more details about the model or framework you are using to assist you better?

Thank you!

4

The model is pretty complex that I don’t really understand, not really sure if it’s helpful. I think it’s an rnn model.

TempUNet(input_dims=self.feature_dims, num_vertices=self.num_vertices,
time_length=self.time_length,
mode=‘nearest’, norm=‘batch’, act_en=‘elu’, act_de=‘leaky_relu’,
en_ksize=self.en_ksize, de_ksize=self.de_ksize,
regularizer_scale=self.regularizer_scale, hidden_dims=self.hidden_dims)

MGConv(self.hidden_dims, {‘XXXX’: self.L_XXXX}, max_diffusion_step, self.num_vertices,
self.regularizer_scale, lambda_max=lambda_max)

XXXX here is npz file.

5

@Lara,

Could you please share details of framework or library that you are using?

Thank you!

6

import numpy as np
import tensorflow as tf
import scipy.sparse as sp
from scipy.sparse import linalg
from tensorflow import keras

7

one of the py file is called unet_model, maybe the user used keras unet?

9

@Lara,

AFAIU, en_ksize and de_ksize are the kernel sizes used in the encoder and decoder of the TempUNet model, where as max_diffusion_step is the maximum number of diffusion steps for message passing in the graph convolution.

Thank you!

1 Like
12

The number of layers is 3, and my en_ksize and de_ksize are both [3,3,3]. What does [3,3,3] mean and look like? Are they related to my number of layers? Can change my de_ksize to [4,7,9] or [1,2,3,4,5]? Thank you so much