Model doesn't seem to learn

Question

I used this gym library to try and get this model to learn, but I don't think it learns from experience. Something is wrong, but I can't figure it out.

I have played with the DISCOUNT, LEARNING_RATE, DISCRETE_OS_SIZE and still nothing, do i have to create a neural network for this example? Or can I just use the formula to derive the q values?

import gym
import numpy as np

LEARNING_RATE = 0.1 
DISCOUNT = 0.95
EPISODES = 25000



env = gym.make("MountainCar-v0")




DISCRETE_OS_SIZE = [20, 20]
discrete_os_win_size = (env.observation_space.high - env.observation_space.low)/DISCRETE_OS_SIZE
q_tables = np.random.uniform(low = -2, high = 0, size = (DISCRETE_OS_SIZE + [env.action_space.n]))

def get_discrete_state(state):
    discrete_state = (state - env.observation_space.low)/DISCRETE_OS_SIZE
    return tuple(discrete_state.astype(np.int))  # we use this tuple to look up the 3 Q values for the available actions in the q-table

# Exploration settings
epsilon = 1  # not a constant, qoing to be decayed
START_EPSILON_DECAYING = 1
END_EPSILON_DECAYING = EPISODES//2
epsilon_decay_value = epsilon/(END_EPSILON_DECAYING - START_EPSILON_DECAYING)


SHOW_EVERY = 1000
done = False

for episode in range(EPISODES):
    discrete_state = get_discrete_state(env.reset())
    done = False

    if episode % SHOW_EVERY == 0:
        render = True
        print(episode)
    else:
        render = False


    while not done:
        if np.random.random() > epsilon:
            # Get action from Q table
            action = np.argmax(q_tables[discrete_state])
        else:
            # Get random action
            action = np.random.randint(0, env.action_space.n)

        new_state, reward, done, _ = env.step(action)#state = position and velocit
        new_discrete_state = get_discrete_state(new_state) 

        if render:
            env.render()

        if not done:
            max_future_q = np.max(q_tables[new_discrete_state]) # Maximum possible Q value in next step (for new state)
            current_q = q_tables[discrete_state + (action,)]# Current Q value (for current state and performed action)
            new_q = (1 - LEARNING_RATE) * current_q + LEARNING_RATE * (reward + DISCOUNT * max_future_q)# And here's our equation for a new Q value for current state and action
            q_tables[discrete_state + (action,)] = new_q# Update Q table with new Q value
              # Simulation ended (for any reson) - if goal position is achived - update Q value with reward directly
        elif new_state[0] >= env.goal_position:
            q_tables[discrete_state + (action,)] = 0
            print(episode)
        discrete_state = new_discrete_state

        # Decaying is being done every episode if episode number is within decaying range
    if END_EPSILON_DECAYING >= episode >= START_EPSILON_DECAYING:
        epsilon -= epsilon_decay_value



env.close()

Answer 1

I have found what the problem was.

In this line discrete_state = (state - env.observation_space.low)/DISCRETE_OS_SIZE I am dividing the (current state - env.observation_space.low) by DISCRETE_OS_SIZE. I came to the conclusion that I am better of dividing by discrete_os_win_size, which solved my problem.