ValueError: Passed in object <KerasVariable shape=(1, 64), dtype=float64, path=dense_15/kernel> of type 'Variable', not tf.Tensor or tf.Variable or ExtensionType.

[hgaur0007]

Hello Everyone,

Could you pleas help me out with the error display above while running the following code:
My Keras version is '3.3.3.dev2024060803'

import tensorflow as tf
import numpy as np
import time
from tensorflow.keras.layers import Dense
import tensorflow_probability as tfp

tf.random.set_seed(42)

class model (tf.keras.Model):
def init(self, layers, train_op, num_epoch, print_epoch):
super(model, self).init()
self.model_layers = layers
self.train_op = train_op
self.num_epoch = num_epoch
self.print_epoch = print_epoch
self.adam_loss_hist = []

def call(self, r):
    return self.r_val(r)

# Running the model    
def r_val(self, r):
   
    r = 2.0*(r - self.bounds["lb"])/(self.bounds["ub"] - self.bounds["lb"]) - 1.0
    for l in self.model_layers:
        r = l(r)
    return r 

def Stress(self, r):
    stress= 448.4*tf.exp(0.1231*(self.r_val(r))) - 0.409*tf.exp(-6.933*(self.r_val(r)))

    return stress
                           
    
# Return the first derivative
def Strain(self, r):
    strain = 0.009312499803*self.r_val(r) +0.009312500112
    return strain
    
                           
def Strain_Energy(self, r, N, a):
    def f(r):                               
        Energy = 0.03*0.015*1*(self.Stress(self.r_val(r)))*(self.Strain(self.r_val(r)))
        return Energy
    value = tf.constant(0, dtype=tf.float64)
    value2 = tf.constant(0, dtype=tf.float64)
    for N in range (1, N+1):            
        value += f(a+((N-(1/2))* ((r-a)/N))) 
    value2= ((r-a)/N)*value
    return value2


def RHS(self, r):
    Ext_En = ((3*0.03*0.015)/(4*184128.9683))*(self.Stress(self.r_val(r)))**2
    return Ext_En

    
#Custom loss function
def get_loss(self, r):
    LHS_Val = self.Strain_Energy(r, 2000, -1)
    RHS_Val = self.RHS(r)
    int_loss = tf.reduce_mean(tf.math.square(LHS_Val - RHS_Val))
    return int_loss
  
# get gradients
def get_grad(self, r):
    with tf.GradientTape() as tape:
        tape.watch(self.trainable_variables)
        L = self.get_loss(r)
    g = tape.gradient(L, self.trainable_variables)
    return L, g
  
# perform gradient descent
def network_learn(self, r):
    self.bounds = {"lb" : tf.math.reduce_min(r),
                   "ub" : tf.math.reduce_max(r)}
    for i in range(self.num_epoch):
        L, g = self.get_grad(r)
        self.train_op.apply_gradients(zip(g, self.trainable_variables))
        self.adam_loss_hist.append(L)
        if i%self.print_epoch==0:
            print("Epoch {} loss: {}".format(i, L))

rmin = -1
rmax = -0.7
numPts = 25
data_type = "float64"
a = tf.constant(-1, dtype=tf.float64)
rint = np.linspace(rmin, rmax, numPts).astype(data_type)
rint = np.array(rint)[np.newaxis].T

#define the model
tf.keras.backend.set_floatx(data_type)
l1 = tf.keras.layers.Dense(25, "tanh")
l2 = tf.keras.layers.Dense(25, "tanh")
l3 = tf.keras.layers.Dense(1, None)
train_op = tf.keras.optimizers.Adam()
num_epoch = 200
print_epoch = 10
pred_model = model([l1, l2, l3], train_op, num_epoch, print_epoch)

#convert the training data to tensors
rint_tf = tf.convert_to_tensor(rint)

#training
print("Training (ADAM)...")
t0 = time.time()
pred_model.network_learn(rint_tf)
t1 = time.time()
print("Time taken (ADAM)", t1-t0, "seconds")

Following Error Appears:

ValueError Traceback (most recent call last)
Cell In[46], line 4
2 print("Training (ADAM)...")
3 t0 = time.time()
----> 4 pred_model.network_learn(rint_tf)
5 t1 = time.time()
6 print("Time taken (ADAM)", t1-t0, "seconds")

Cell In[42], line 83, in model.network_learn(self, r)
80 self.bounds = {"lb" : tf.math.reduce_min(r),
81 "ub" : tf.math.reduce_max(r)}
82 for i in range(self.num_epoch):
---> 83 L, g = self.get_grad(r)
84 self.train_op.apply_gradients(zip(g, self.trainable_variables))
85 self.adam_loss_hist.append(L)

Cell In[42], line 73, in model.get_grad(self, r)
71 def get_grad(self, r):
72 with tf.GradientTape() as tape:
---> 73 tape.watch(self.trainable_variables)
74 L = self.get_loss(r)
75 g = tape.gradient(L, self.trainable_variables)

File ~\AppData\Roaming\Python\Python311\site-packages\tensorflow\python\eager\backprop.py:873, in GradientTape.watch(self, tensor)
864 def watch(self, tensor):
865 """Ensures that tensor is being traced by this tape.
866
867 Args:
(...)
871 ValueError: if it encounters something that is not a tensor.
872 """
--> 873 for t in _extract_tensors_and_variables(tensor):
874 if not backprop_util.IsTrainable(t):
875 logging.log_first_n(
876 logging.WARN, "The dtype of the watched tensor must be "
877 "floating (e.g. tf.float32), got %r", 5, t.dtype)

File ~\AppData\Roaming\Python\Python311\site-packages\tensorflow\python\eager\backprop.py:700, in _extract_tensors_and_variables(tensor)
698 yield from _extract_tensors_and_variables(components)
699 else:
--> 700 raise ValueError(f"Passed in object {obj} of type {type(obj).name!r}"
701 f", not tf.Tensor or tf.Variable or ExtensionType.")

ValueError: Passed in object <KerasVariable shape=(1, 64), dtype=float64, path=dense_15/kernel> of type 'Variable', not tf.Tensor or tf.Variable or ExtensionType.