def do_spin(self, spin_angle):
timer = 0
final_time = math.fabs(spin_angle) / self.spin_speed
self.twist_cmd.angular.z = utils.sgn(spin_angle) * self.spin_speed
while timer < final_time:
self.twist_commander.publish(self.twist_cmd)
timer += self.sample_dt
self.loop_timer.sleep()
self.do_halt()
def load_patterns(n_patterns: Optional[int] = None) -> np.ndarray:
"""
Returns training images from the mnist dataset in hopfield-network form.
More specifically, 8-bit images from (0, 255) are converted to
floating point images having values in {-1, 1}.
"""
mnist = load_mnist()
out = mnist["images"][:n_patterns]
out = 2 * (out.astype(np.float32) / 255.0) - 1
out = sgn(out)
return out
def learn_aen(self, x, ir):
b_new = self.B + self.beta * ir * self.x_prev
c_new = self.C + self.beta * ir * self.y_prev
a_new = np.zeros_like(self.A)
for i in range(self.h_aen):
for j in range(self.input_len + 1):
a_new[j,
i] = self.A[j, i] + self.beta_h * ir * self.y_prev[i] * (
1 - self.y_prev[i]) * sgn(self.C[i]) * self.x_prev[j]
self.A = a_new
self.B = b_new
self.C = c_new
def do_move(self, distance):
timer = 0
final_time = math.fabs(distance) / self.move_speed
self.twist_cmd.angular.z = 0
self.twist_cmd.linear.x = utils.sgn(distance) * self.move_speed
while timer < final_time:
self.twist_commander.publish(self.twist_cmd)
timer += self.sample_dt
self.loop_timer.sleep()
self.do_halt()
def learn_asn(self, x, ir, s):
e_new = self.E + self.rho * ir * s * self.x_prev
f_new = self.F + self.rho * ir * s * self.z_prev
d_new = np.zeros_like(self.D)
for i in range(self.h_asn):
for j in range(self.input_len + 1):
d_new[j,
i] = self.D[j, i] + self.rho_h * ir * self.z_prev[i] * (
1 - self.z_prev[i]) * sgn(
self.F[i]) * s * self.x_prev[j]
self.D = d_new
self.E = e_new
self.F = f_new
def derivative(self, y): # computed yd
g = 9.81 # gravity constant
# Compute thetadd by splitting the computation of the numerator and denominator for clarity
temp_num = g * sin(y[2]) + cos(y[2]) * (
-self.f - self.m * self.l * y[3] * y[3] * sin(y[2]) + self.mu_c * sgn(y[1])) / (
self.mu_c + self.m) - self.mu_p * y[3] / (self.m * self.l)
temp_denom = self.l * (4 / 3 - (self.m * cos(y[2]) ** 2) / (self.mu_c + self.m))
thetadd = temp_num / temp_denom
# Compute xdd by splitting the computation of the numerator and denominator for clarity
temp_num = self.f + self.m * self.l * (y[3] ** 2 * sin(y[2]) - thetadd * cos(y[2])) - self.mu_c * sgn(y[1])
temp_denom = self.mu_c + self.m
xdd = temp_num / temp_denom
return np.array([y[1], xdd, y[3], thetadd])
def update(self) -> None:
# Compute synaptic input.
self._I_syn = np.dot(self._weights, self._state)
self._state = sgn(self._bias + self._I_syn)
def k_func_cart_pole(v, v_prev, p):
return 1 - p if sgn(v) != sgn(v_prev) else -p
def _choose_determinant_action(self, meadow, directions, all_bees):
if self.pollen_gathered >= self.max_capacity:
return (sgn(self.hive_position[0] - self.position[0]), sgn(self.hive_position[1] - self.position[1]))
else:
return self._neural_move(meadow, directions, all_bees)