total += 1
loss = loss / total
display_data["loss"] = loss
history["loss"].append(loss)
total = 0
correct = 0
for x_v_batch, y_v_batch in itertools.islice(generate_batch(x_valid), None, len(x_valid) // config.batch_size):
prediction = model.classify(x_v_batch)
for j in range(len(prediction)):
total += 1
if prediction[j] == y_v_batch[j]:
correct += 1
acc = correct / total
display_data["eval_acc"] = acc
history["acc"].append(acc)
bar.set_postfix(display_data)
model.save_weights(weights_dir, i)
model.save(model_dir)
utils.save_class_ids(activity_dict, os.path.join(model_dir, ACTIVITY_ID_NAME_MAPPING_FILENAME))
index = np.argmax(history['acc'])
comment = "Max accuracy {} after {} epoch".format(history['acc'][index], index)
print(comment)
plot_utils.create_plot(history['acc'], history['loss'], os.path.join(model_dir, "results.png"), True, comment)
gamma = 0.98
#
alpha = 0.1
nr_episodes = 2_000
epsilon = 0.4
# 1.2 policy $\pi$
# initializing policy to a random policy
# initializing Q to zero
pi = return_a_random_policy(n, nr_actions)
Q = np.zeros((n, n, nr_actions))
# 1.6 setting up the plot
ax = create_plot(n)
plt.ion()
interactive(True)
plt.cla()
ax.axis('off')
for episode_id in tqdm(range(nr_episodes)):
terminated = False
state = initialize_the_state(n)
while not terminated:
pi = return_epsilon_greedy_pi(Q, epsilon)
action_id = choose_an_action_based_on_pi(state, pi)
new_state, reward, terminated = step(state, action_id, n)
learn_Q(state, action_id, reward, new_state, Q, gamma, alpha)
state = new_state + 0.0
np.random.seed(0)
# 1.2 grid size along each direction
N = 8
env = Maze(N=N, wall_length=2)
# 1.2 policy $\pi$
# initializing policy to a random policy
# initializing V to zero
pi = return_a_random_policy(N, env.action_space.n, epsilon=1000000)
V_accumulate = np.zeros((N, N))
# 1.6 setting up the plot
ax = create_plot(N)
plt.ion()
interactive(True)
plt.cla()
ax.axis('off')
nr_episodes = 1_000
gamma = 0.98
all_states = dstack_product(np.arange(N), np.arange(N))
for episode_id in tqdm(range(nr_episodes)):
# a sweep over all the states in the system.
for counter, init_state in enumerate(all_states):
terminated = False
env.reset(init_state)
def plot():
graphs= create_plot()
return render_template('plot.html', plot0=graphs[0], plot1=graphs[1],plot2=graphs[2],plot3=graphs[3],plot4=graphs[4] )