drift_mu=np.zeros(3),
drift_sd=1.)
data = generate_behavioural_data(MyBanditTask, RWSoftmaxAgent, 20, 200)
def log_prob(w, D):
agent = RWSoftmaxAgent(task=MyBanditTask(),
learning_rate=w[0],
inverse_softmax_temp=w[1])
L = 0
for t in range(D.shape[0]):
x = D[t, :7]
u = D[t, 7:11]
r = D[t, 11]
x_ = D[t, 12:]
L += u @ agent.log_prob(x)
agent.learning(x, u, r, x_, None)
return L
res = mlepar(log_prob, data.tensor, 2, maxstarts=5)
X = res.transform_xmin([sigmoid, relu])
# Criticism: Actual vs. Estimate Plots
lr_fig = actual_estimate(data.params[:, 1], X[:, 0])
plt.show()
ist_fig = actual_estimate(data.params[:, 2], X[:, 1])
plt.show()
def test_actual_estimate():
x = np.linspace(0, 10, 20)
f = actual_estimate(x, x, xlabel='x', ylabel='y')
del (f)
plt.close()
data = generate_behavioural_data(TwoArmedBandit, RWSoftmaxAgent, N, T)
# Create log-likelihood function
def log_prob(w, D):
lr = sigmoid(w[0], a_min=-6, a_max=6)
ist = stable_exp(w[1], a_min=-10, a_max=10)
agent = RWSoftmaxAgent(TwoArmedBandit(), lr, ist)
L = 0
for t in range(D.shape[0]):
x = D[t, :3]
u = D[t, 3:5]
r = D[t, 5]
x_ = D[t, 6:]
agent.log_prob(x, u)
agent.learning(x, u, r, x_, None)
J = np.array([grad.sigmoid(w[0]), grad.exp(w[1])])
return -agent.logprob_, -J * agent.grad_,
# Fit model
res = mlepar(log_prob, data.tensor, nparams=2, maxstarts=5, jac=True)
X = res.transform_xmin([sigmoid, stable_exp])
idx = np.logical_and(np.logical_not(np.isnan(X[:, 0])), np.less(X[:, 1], 20))
# Criticism: Actual vs. Estimate Plots
lr_fig = actual_estimate(data.params[idx, 1], X[idx, 0])
plt.show()
ist_fig = actual_estimate(data.params[idx, 2], X[idx, 1])
plt.show()
maxstarts=4,
maxstarts_without_improvement=2,
init_sd=1,
njobs=-1,
jac=True,
hess=True,
method='trust-exact')
res_rwssm = mlepar(f=rwstickysoftmax_loglik,
data=data_rwsticky.tensor,
nparams=3,
minstarts=2,
maxstarts=4,
maxstarts_without_improvement=2,
init_sd=1,
njobs=-1,
jac=True,
hess=True,
method='trust-exact')
res.xmin
xhat = res.xmin[np.logical_not(np.any(np.isnan(res.xmin), axis=1)), :]
xhat = np.stack(
fu.transform(xhat[i], [fu.sigmoid, fu.stable_exp]).flatten()
for i in range(xhat.shape[0]))
xtrue = data.params[np.logical_not(np.any(np.isnan(res.xmin), axis=1)), 1:]
f = actual_estimate(xtrue[:, 0], xhat[:, 0])
f = actual_estimate(xtrue[xhat[:, 1] < 20, 1], xhat[xhat[:, 1] < 20, 1])
q.learning(X1[t], U1[t], R[t], X2[t], u2)
L = q.logprob_
g = q.grad_[[0, 1, 3]]
H = np.hstack((q.hess_[:, :2], q.hess_[:, -1].reshape(-1, 1)))
H = np.vstack((H[:2, :], H[-1, :].reshape(1, -1)))
return -L, -J @ g, -J.T @ H @ J
res = mlepar(f=loglik,
data=data.tensor,
nparams=3,
minstarts=2,
maxstarts=10,
maxstarts_without_improvement=2,
init_sd=1,
njobs=-1,
jac=True,
hess=True,
method='trust-exact')
xhat = res.xmin[np.logical_not(np.any(np.isnan(res.xmin), axis=1)), :]
xhat = np.stack(
fu.transform(xhat[i], [fu.sigmoid, fu.stable_exp, fu.sigmoid]).flatten()
for i in range(xhat.shape[0]))
xtrue = data.params[np.logical_not(np.any(np.isnan(res.xmin), axis=1)), 1:]
f = actual_estimate(xtrue[:, 0], xhat[:, 0])
f = actual_estimate(xtrue[xhat[:, 3] < 20, 3], xhat[xhat[:, 3] < 20, 3])
f = actual_estimate(xtrue[xhat[:, 3] < 20, 1], xhat[xhat[:, 3] < 20, 1])
f = actual_estimate(xtrue[xhat[:, 3] < 20, 2], xhat[xhat[:, 3] < 20, 2])