def test_error_when_trying_to_plot_from_Network_show_normal_modes(self):
circuit = core.Network([
core.C(0,1,'C'),
core.J(0,1,'Lj')
])
with self.assertRaises(TypeError):
circuit.show_normal_mode()
def test_q_zpf(self):
Cj = 100e-15
Lj = 10e-9
junction = core.J(0, 1, Lj)
circuit = core.Network([core.C(0, 1, Cj), junction, core.R(0, 1, 1e6)])
Z = np.sqrt(Lj / Cj)
q_zpf = np.sqrt(hbar / Z / 2)
self.assertRelativelyClose(
q_zpf / e, np.absolute(junction.zpf(mode=0, quantity='charge')))
def parameters(self, Cj, Lj, Cc, Cr, Lr):
circuit = core.Network([
core.C(0, 1, Cj),
core.J(0, 1, Lj),
core.C(1, 2, Cc),
core.C(0, 2, Cr),
core.L(0, 2, Lr)
])
return circuit.f_k_A_chi()
def test_anharmonicity_using_hamiltonian(self):
Cj = 1e-10
circuit = core.Network([core.C(0, 1, Cj), core.J(0, 1, 10e-9)])
H = circuit.hamiltonian(modes=[0], taylor=4, excitations=[10])
ee = H.eigenenergies()
A = np.absolute((ee[1] - ee[0]) - (ee[2] - ee[1]))
# Due to higher order terms, the mismatch with e**2/2/Cj/h is
# (193702.3+0j) != (194712.7+0j)
A_expected = 194712.7
self.assertRelativelyClose(A_expected, A)
def test_double_series_capacitor(self):
C = 100e-15
Lj = 10e-9
circuit = core.Network(
[core.C(0, 1, C * 2),
core.C(1, 2, C * 2),
core.J(0, 2, Lj)])
f, k, A, chi = circuit.f_k_A_chi()
self.assertArrayRelativelyClose(
[e**2 / 2. / C / h, 1 / (np.sqrt(C * Lj) * 2. * pi)], [A[0], f[0]])
def parameters(self, C, L):
circuit = core.Network([
core.C(0, 2, C),
core.C(1, 3, C),
core.J(0, 1, L),
core.J(1, 2, L),
core.J(2, 3, L),
core.J(3, 0, L)
])
return circuit.f_k_A_chi()
def test_sweeping_LJ_in_fkAchi(self):
cir = core.Network([
core.C(0, 1, 100e-15),
core.J(0, 1, 'L_J'),
core.C(1, 2, 1e-15),
core.C(2, 0, 100e-15),
core.L(2, 0, 10e-9),
core.R(2, 0, 1e6)
])
[cir.f_k_A_chi(L_J=x) for x in [1e-9, 2e-9]]
def test_phi_zpf(self):
Cj = 100e-15
Lj = 10e-9
junction = core.J(0, 1, Lj)
circuit = core.Network([core.C(0, 1, Cj), junction, core.R(0, 1, 1e6)])
phi_0 = hbar / 2 / e
Z = np.sqrt(Lj / Cj)
phi_zpf = np.sqrt(hbar * Z / 2)
self.assertRelativelyClose(phi_zpf / phi_0,
junction.zpf(mode=0, quantity='flux'))
def test_LC_double_series_L_double_series_C(self):
C = 1e-8
L = 3
circuit = core.Network([
core.C(0, 1, C * 2),
core.C(1, 2, C * 2),
core.L(2, 3, L / 2),
core.L(3, 0, L / 2)
])
f, k, A, chi = circuit.f_k_A_chi()
f_expected = 1 / np.sqrt(L * C) / 2 / np.pi
self.assertRelativelyClose(f_expected, f)
def test_sweeping_CJ_array_in_zpf(self):
C_comp = core.C(0, 1, 'C_J')
cir = core.Network([
C_comp,
core.J(0, 1, 10e-9),
core.C(1, 2, 1e-15),
core.C(2, 0, 100e-15),
core.L(2, 0, 10e-9),
core.R(2, 0, 1e6)
])
self.assertRelativelyClose(
C_comp.zpf(mode=1, quantity='charge', C_J=1.5e-9),
C_comp.zpf(mode=1, quantity='charge', C_J=[1e-9, 1.5e-9, 3e-9])[1])
def revaluing_labelled_valued_component_twice(self):
'''
Adressing last error appearing in issue #83
'''
cir = core.Network(
[core.L(0, 1, 1),
core.C(0, 1, 1),
core.R(0, 1, 'R', 1)])
try:
cir.loss_rates(R=1)
except Exception:
pass
with self.assertRaises(ValueError):
cir.loss_rates(R=1)
def parameters(self, R, L, C):
circuit = core.Network(
[core.C(0, 1, C),
core.L(1, 2, L),
core.R(0, 2, R)])
return circuit.f_k_A_chi()
def parameters(self, C, Lj):
circuit = core.Network([core.C(0, 1, C), core.J(0, 1, Lj)])
return circuit.f_k_A_chi()
plt.show()
plt.clf()
# CartPole-v0
if example == "CartPole":
def score(rewards):
return np.sum(rewards)
env = gym.make('CartPole-v0')
net = pf.Network(network={
'type': 'fc',
'layers': [4, 64, 64, 2]
},
agent='reinforce',
actions=[0, 1],
learning_rate=0.0001,
decay_steps=1000,
decay_factor=0.90,
keep_prob=1.00)
agent = pf.Reinforce(net=net,
env=env,
discount_rate=1.00,
max_episodes=500,
score=score)
agent.train()
evaluate(env, agent, score, 100)
# LunarLander-v1
if example == "LunarLander":
def test_error_when_trying_to_plot_from_Network_show(self):
circuit = core.Network([core.C(0, 1, "C"), core.J(0, 1, "Lj")])
with self.assertRaises(TypeError):
circuit.show()
import tensorflow as tf
import numpy as np
import core as pf
import time
import gym
import sys
env = gym.make('CartPole-v0')
for lr in 10.0**np.random.uniform(-3, -5, size=10):
print("Training with lr={:f}".format(lr))
net = pf.Network(network={
'type': 'fc',
'layers': [4 * 1, 256, 256, 2]
},
agent='deepq',
actions=[0, 1],
learning_rate=lr,
decay_steps=1000,
decay_factor=0.99,
keep_prob=0.50,
batch_size=32)
net.build()
agent = pf.DeepQ(
net=net,
env=env,
discount_rate=0.99,
max_episodes=500,
score=np.sum,
min_mem=10**4, # 5-10% of max_mem
max_mem=10**5,
seq_length=1,
import tensorflow as tf
import numpy as np
import core as pf
import time
import gym
import sys
env = gym.make('CartPole-v0')
net = pf.Network(network={
'type': 'fc',
'layers': [4, 1024, 512, 2]
},
agent='reinforce',
actions=[0, 1],
learning_rate=0.0001,
decay_steps=1000,
decay_factor=0.90,
keep_prob=0.50)
net.build()
agent = pf.Reinforce(net=net,
env=env,
discount_rate=1.00,
max_episodes=500,
print_episodes=10,
score=np.sum)
agent.train()
agent.evaluate()
min_epsilon = 0.01
max_episodes = 500
print_episodes = 10
def score(x):
return np.sum(x * 0.99**np.range(1, len(x) + 1))
env = gym.make('LunarLander-v2')
net = pf.Network(network={
'type': 'fc',
'layers': [8, 256, 256, 4]
},
agent='deepq',
actions=[0, 1, 2, 3],
learning_rate=learning_rate,
decay_steps=lr_decay_steps,
decay_factor=lr_decay_factor,
keep_prob=keep_prob,
batch_size=batch_size)
net.build()
agent = pf.DeepQ(net=net,
env=env,
discount_rate=discount_rate,
max_episodes=max_episodes,
score=score,
min_mem=min_mem,
max_mem=max_mem,
seq_length=seq_length,
batch_size=batch_size,
