model returning a tensor with values 0 occasionally with no schange in input

[champion of wits]

Hi

I was trying to build  a simple auto encoder , while executing the following code snippet i was getting either a full zero filled array , or proper output despite not changing the code.

I have tested this code on my local system and google co laboratory ,both returning same issue

code snippet:

import tensorflow as tf

from tensorflow import keras

from tensorflow.keras import layers

import scipy

import numpy as np

import cv2

tf.compat.v1.enable_eager_execution()

def generate_data_set():

     l = []

     for i in range(50):

         l.append([i,i+2,i+4])

     return np.asarray(l ,dtype = np.single )

m = generate_data_set()

#print(m)

model = keras.Sequential(

[

layers.Dense(9,activation = "relu" , name ="encode_1"),

layers.Dense(5,activation = "relu" , name ="encode_2"),

layers.Dense(1,activation = "relu" , name ="shorty"),

layers.Dense(5,activation = "relu" , name = "decode_2"),

layers.Dense(9,activation = "relu" , name = "decode_1"),

layers.Dense(3,name = "out"),

]

)

x = tf.ones((3,3))

y = model(x)

print(y.numpy())