def main():
from environments import MultiArmedBandit
import seaborn as sns
sns.set(style="white", palette="muted", color_codes=True)
T = 100
env = MultiArmedBandit(T)
pm = RescorlaWagner(env, env.d_x)
obs = env.get_observations()
hst = env.get_hidden_states()
#use isres for optimisation of one dimensional functions
from optmethods import isres
bounds = {'ub': np.array([1.]), 'lb': np.array([0.])}
f_opt, x_opt, res = isres(pm.get_free_energy, 1, 1e-6, 1e-8, bounds,
np.array([0.5]))
print(f_opt, x_opt, res)
post = pm.get_beliefs(alpha=x_opt)
ax = obs.plot(y=r'$o_t$', style='go')
ax = hst.plot(y=r'$p_t$', style='k--', ax=ax)
ax = post.plot(y=r'$\mu_t$', style='r-', ax=ax)
ax.legend(numpoints=1)
#optimize preceptual surprise over multiple experimental blocks
def total_fe(x, n_pars, blocks):
fe = 0
for b in blocks:
pm = RescorlaWagner(b, b.d_x)
fe += pm.get_free_energy(x)
return fe
n = 100
T = 100
exp_blocks = [MultiArmedBandit(T)] * 100
fe = lambda x, p: total_fe(x, p, exp_blocks)
f_opt, x_opt, res = isres(fe, 1, 1e-6, 1e-8, bounds, np.array([0.5]))
print(f_opt / n, x_opt, res)
post = pm.get_beliefs(alpha=x_opt)
ax = obs.plot(y=r'$o_t$', style='go')
ax = hst.plot(y=r'$p_t$', style='k--', ax=ax)
ax = post.plot(y=r'$\mu_t$', style='r-', ax=ax)
ax.legend(numpoints=1)
def main():
from environments import MultiArmedBandit
import seaborn as sns
sns.set(style = "white", palette="muted", color_codes=True)
T = 100
env = MultiArmedBandit(T)
pm = RescorlaWagner(env, env.d_x)
obs = env.get_observations()
hst = env.get_hidden_states()
#use isres for optimisation of one dimensional functions
from optmethods import isres
bounds = {'ub': np.array([1.]), 'lb': np.array([0.])}
f_opt, x_opt, res = isres( pm.get_free_energy, 1, 1e-6, 1e-8, bounds, np.array([0.5]) )
print(f_opt, x_opt, res)
post = pm.get_beliefs(alpha = x_opt)
ax = obs.plot(y = r'$o_t$', style = 'go')
ax = hst.plot(y = r'$p_t$', style = 'k--', ax = ax)
ax = post.plot(y = r'$\mu_t$', style = 'r-', ax = ax)
ax.legend(numpoints = 1)
#optimize preceptual surprise over multiple experimental blocks
def total_fe(x, n_pars, blocks):
fe = 0
for b in blocks:
pm = RescorlaWagner(b, b.d_x)
fe += pm.get_free_energy(x)
return fe
n = 100
T = 100
exp_blocks = [MultiArmedBandit(T)]*100
fe = lambda x,p: total_fe(x, p, exp_blocks)
f_opt, x_opt, res = isres( fe, 1, 1e-6, 1e-8, bounds, np.array([0.5]) )
print(f_opt/n, x_opt, res)
post = pm.get_beliefs(alpha = x_opt)
ax = obs.plot(y = r'$o_t$', style = 'go')
ax = hst.plot(y = r'$p_t$', style = 'k--', ax = ax)
ax = post.plot(y = r'$\mu_t$', style = 'r-', ax = ax)
ax.legend(numpoints = 1)
