def policy_rollout(GP_model, depth_h,bounds):
# caluc policy function
if depth_h > 1:
func2minimize = lambda x : -1*ei(x,bounds,GP_model)
else:
func2minimize = lambda x : gpmean(x,bounds,GP_model)
query = minimize(func2minimize, bounds)
return query
def policy_rollout_ei(GP_model, func_cost, depth_c, threthold, bounds):
# caluc policy function
if depth_c > threthold:
func2minimize = lambda x: -1 * ei(x, bounds, GP_model)
else:
func2minimize = lambda x: gpmean(x, bounds, GP_model)
query = minimize(func2minimize, bounds)
return query
content_target = K.variable(content_net.predict(content_image))
# 7 - Calculate the loss and gradients respect to input
# 7.1 - Content loss
loss = K.mean(K.square(content_net.output - content_target))
grads = K.gradients(loss, content_net.input)
# 7.4 - Create a callable function that uses the symbolic variables we have created
loss_grads = K.function(inputs=[content_net.input], outputs=[loss] + grads)
# 7.5 - Wrapper function
def get_loss_and_grads_wrapper(x_vec):
'''
Takes a 1D vector, reshape it into an image, calculate loss and gradient
and flatten each vector back casting into float64
'''
l, g = loss_grads([x_vec.reshape(*batch_shape)])
return l.astype(np.float64), g.flatten().astype(np.float64)
final_image, history = minimize(get_loss_and_grads_wrapper, 10, batch_shape)
plt.imshow(scale_(final_image))
plt.savefig(os.path.join(path_to_save, 'content_decoded.png'))
plt.show()
plt.plot(history)
plt.savefig('history.png')
plt.show()
kB = unit.BOLTZMANN_CONSTANT_kB * unit.AVOGADRO_CONSTANT_NA
density = reduced_density / (sigma**3)
temperature = reduced_temperature * epsilon / kB
pressure = reduced_pressure * epsilon / (sigma**3)
kT = kB * temperature
# Create the Lennard-Jones fluid.
testsystem = testsystems.LennardJonesFluid(nparticles=nparticles,
mass=mass,
sigma=sigma,
epsilon=epsilon,
reduced_density=reduced_density)
# Construct initial positions by minimization.
print "Minimizing positions..."
initial_positions = utils.minimize(testsystem.system, testsystem.positions)
# Write initial positions.
print "Writing initial positions to initial.pdb"
utils.write_pdb('initial.pdb', initial_positions)
# Create NetCDF file to store data.
ncfile = utils.create_netcdf_datastore(netcdf_filename, testsystem.system,
initial_positions, nreplicates,
niterations, observation_interval)
# Run replicates of the simulation.
for replicate in range(nreplicates):
# Reconstitute System object.
print "Reconstituting System object..."
system = openmm.System()
def policy_naive(GP_model, depth_h,bounds):
# caluc naive policy function
func2minimize = lambda x : -1*ei(x,bounds,GP_model)
query = minimize(func2minimize, bounds)
return query
def policy_mu(GP_model, bounds):
# caluc naive policy function
func2minimize = lambda x: gpmean(x, bounds, GP_model)
query = minimize(func2minimize, bounds)
return query
def policy_naive_eipercost(GP_model, func_cost, depth_c, threthold, bounds):
# caluc naive policy function
func2minimize = lambda x: -1 * ei_per_cost(x, func_cost, bounds, GP_model)
query = minimize(func2minimize, bounds)
return query
nsteps_per_iteration = barostat_frequency
observation_interval = timestep * nsteps_per_iteration
# Compute real units.
kB = unit.BOLTZMANN_CONSTANT_kB * unit.AVOGADRO_CONSTANT_NA
density = reduced_density / (sigma**3)
temperature = reduced_temperature * epsilon / kB
pressure = reduced_pressure * epsilon / (sigma**3)
kT = kB * temperature
# Create the Lennard-Jones fluid.
testsystem = testsystems.LennardJonesFluid(nparticles=nparticles, mass=mass, sigma=sigma, epsilon=epsilon, reduced_density=reduced_density)
# Construct initial positions by minimization.
print "Minimizing positions..."
initial_positions = utils.minimize(testsystem.system, testsystem.positions)
# Write initial positions.
print "Writing initial positions to initial.pdb"
utils.write_pdb('initial.pdb', initial_positions)
# Create NetCDF file to store data.
ncfile = utils.create_netcdf_datastore(netcdf_filename, testsystem.system, initial_positions, nreplicates, niterations, observation_interval)
# Run replicates of the simulation.
for replicate in range(nreplicates):
# Reconstitute System object.
print "Reconstituting System object..."
system = openmm.System()
system.__setstate__(str(ncfile.variables['system'][0]))
