Error: Something unsupported is present in the user code, add help info

I have a python code to accelerate using numba. It involves a class, tuples and dictionary. Can anyone help me find a wayout in this code?
def q_learn(EPSILON,episode_rewards,q_table):

for episode in range(N_EPISODES+1):

    environment = Environment(some integer data)
    print(f'Currently at {episode} Episode')

    if episode%SHOW_EVERY==0:
        print(f'Episode {episode} epsilon:{EPSILON}')
        print(f'{SHOW_EVERY} episode mean reward {np.mean(episode_rewards[-SHOW_EVERY:])}')
        with open(f"q-table_age{int(time.time())}.pickle","wb") as f:

    epi_rew = 0.0
    for i in range(100):
        rd = random.randint(0,15)
        x = test_data[rd]
        y = test_data[rd]
        obs = (tuples of size 3 )
        if np.random.random()>EPSILON:
            action = np.argmax(q_table[obs])+1

            action = np.random.randint(1,4)

        reward = environment.action(3 integer and 1 float arguments)

        new_obs = (tuple of size 3)

        max_future_q = np.max(q_table[new_obs])
        current_q = q_table[new_obs][action-1]

        new_q = (1-LEARNING_RATE)*current_q + LEARNING_RATE*(reward + DISCOUNT*max_future_q)

        q_table[obs][action-1] = new_q


    episode_rewards[episode] = epi_rew

moving_avg = np.convolve(episode_rewards,np.ones((SHOW_EVERY,))/SHOW_EVERY,mode="Valid")

return moving_avg

That’s a pretty broad question… in general, you can’t just decorate an arbitrary function with ‘@njit’ and expect it to work.

I’d suggest working some simple tests, perhaps starting with the typed dict and building incrementally from there.