def dyian_test(path):
path_plot = path + "plot_pt\\"
if not os.path.exists(path_plot):
os.makedirs(path_plot)
path_save = path + "save_pt\\"
if not os.path.exists(path_save):
os.makedirs(path_save)
diayn_mod = build_diayn(2)
diayn_mod.load(path + "save_diayn\\")
mountaincar(diayn_mod, path_plot + "pretrained_trajectoires_0")
pretrained = A2C.from_diayn(diayn_mod, 0)
model = pretrained
for k in range(0, 1):
iter_ = 200
model.train(iter_)
model.plot_rewards(path_plot + "pretrained_rewards_" +
str((k + 1) * iter_))
mountaincar_baseline(
model, path_plot + "pretrained_trajectoires_" + str(
(k + 1) * iter_))
model.save(path_save)
def plot_results(path):
env = gym.make("MountainCar-v0")
baseline = A2C(env, {"actor": [30, 30], "critic": [30, 30]}, gamma=0.99)
baseline.load(path + "save_bl\\")
diayn_mod = build_diayn(2)
diayn_mod.load(path + "save_diayn\\")
pretrained = A2C.from_diayn(diayn_mod, 0)
pretrained.load(path + "save_pt\\")
plt.figure()
plt.plot(range(99, len(pretrained.rewards)),
np.convolve(pretrained.rewards,
np.ones(100) / 100, "valid"),
label="pretrained")
plt.plot(range(99, len(baseline.rewards)),
np.convolve(baseline.rewards,
np.ones(100) / 100, "valid"),
label="baseline")
plt.legend()
plt.show()
plt.savefig(path + "results")
plt.pause(1)