I was trying to train a model on a gym environment when I encountered the following error:

TypeError: Argument `fetch` = <tf.Variable 'dense/kernel:0' shape=(4, 24) dtype=float32> has invalid type "ResourceVariable" must be a string or Tensor. (Can not convert a ResourceVariable into a Tensor or Operation.)

Here is the relevant code:

env = gym.make("CartPole-v1")

env.observation_space.sample()
from tensorflow.python.keras.models import Sequential
from tensorflow.python.keras.layers import Dense, Flatten
from tensorflow.keras.optimizers import Adam  

states = env.observation_space.shape
actions = env.action_space.n

print("Actions: " + str(actions))

def build_model(states, actions):
    model = Sequential()    
    model.add(Dense(24, activation='relu', input_shape=states))
    model.add(Dense(24, activation='relu'))
    model.add(Dense(actions, activation='linear'))
    return model

model = build_model(states, actions)


model.summary()
from rl.agents import DQNAgent
from rl.policy import BoltzmannQPolicy
from rl.memory import SequentialMemory

def build_agent(model, actions):
    policy = BoltzmannQPolicy()
    memory = SequentialMemory(limit=50000, window_length=1)
    dqn = DQNAgent(model=model, memory=memory, policy=policy, 
                  nb_actions=actions, nb_steps_warmup=10, target_model_update=1e-2)
    return dqn


dqn = build_agent(model, actions)

dqn.compile(Adam(lr = 0.001), metrics= ["mae"])

dqn.fit(env, nb_steps=10000, visualize=False, verbose=1)

I tried change my versions to what others say is compatible.

tensorflow: 2.12.0
gym: 0.25.2
keras-rl2: 1.0.5 

I also know this is not a cartpole problem since I was originally trying to get this to work on a custom gym environment which gave the same error. Changing how the optimizer is passed in also does not affect the error.