# agent convert to torch model
import policy_torch
baselines_mlp_model =TD3.load('trained_agent/accac_final_1000000_9_11_20_25.zip')
agent = policy_torch.copy_mlp_weights(baselines_mlp_model,layers=[512,512],n_inputs=32)
# loading enviorment, same as training
env=firefly_accac.FireflyAccAc(arg)
# ---seting the env for inverse----
# TODO, move it to a function of env
env.agent_knows_phi=False
for i in range(10):
filename=(str(time.localtime().tm_mday)+'_'+str(time.localtime().tm_hour)+'_'+str(time.localtime().tm_min))
single_theta_inverse(arg, env, agent, filename,
number_updates=number_updates,
true_theta=None,
phi=None,
init_theta=None,
states=None,
actions=None,
trajectory_data=None,
use_H=False,
tasks=None,
)
print('done')
# [0.15],
# [1],
# [0.03],
# [0.03]]).cuda()
for i in range(1):
filename = (str(time.localtime().tm_mday) + '_' +
str(time.localtime().tm_hour) + '_' +
str(time.localtime().tm_min))
single_theta_inverse(
arg,
env,
agent,
filename,
number_updates=number_updates,
true_theta=None,
phi=None,
init_theta=None,
states=None,
actions=None,
trajectory_data=None,
use_H=False,
tasks=None,
is1d=True,
gpu=False
# fixed_param_ind=[5,6],
# assign_true_param=[5,6],
# task=[torch.tensor([0.7]).cuda()],
)
print('done')
'trained_agent//acc_retrain_1000000_2_18_21_4.zip')
agent = policy_torch.copy_mlp_weights(baselines_mlp_model,
layers=[128, 128],
n_inputs=30)
# loading enviorment, same as training
env = firefly_acc.FireflyAcc(arg)
# ---seting the env for inverse----
# TODO, move it to a function of env
env.agent_knows_phi = False
for i in range(10):
filename = ("test_acc_EP" + str(arg.NUM_EP) + "updates" +
str(number_updates) + "lr" + str(arg.ADAM_LR) + 'step' +
str(arg.LR_STEP) + str(time.localtime().tm_mday) + '_' +
str(time.localtime().tm_hour) + '_' +
str(time.localtime().tm_min))
single_theta_inverse(arg,
env,
agent,
filename,
number_updates=number_updates,
true_theta=None,
phi=None,
init_theta=None,
states=None,
actions=None,
tasks=None)
print('done')
[[0.2207319438457489],
[1.062300205230713],
[0.32934996485710144],
[0.1929050236940384],
[0.19170257449150085],
[0.1894093006849289],
[0.16225792467594147],
[0.05502069741487503],
[0.6376186013221741],
[0.7159334421157837]]
)
for i in range(1):
filename=(str(time.localtime().tm_mday)+'_'+str(time.localtime().tm_hour)+'_'+str(time.localtime().tm_min))
single_theta_inverse(arg, env, agent, filename,
number_updates=number_updates,
true_theta=None,
phi=phi,
init_theta=theta,
trajectory_data=(states, actions, tasks),
use_H=False,
is1d=False,
gpu=False,
# fixed_param_ind=[1,2,5,6],
# assign_true_param=[1,2,5,6],
action_var=0.001
)
