module 'baselines.deepq' has no attribute 'load'

[ChenyangRan]

Hi, I am new to pybullet and baselines, so sorry, if I am asking a unnecessary question.
When I run the enjoy_kuka_grasping.py, some errors happens.
It shows that ### module 'baselines.deepq' has no attribute 'load'.
Maybe is

act = deepq.load("kuka_model.pkl")

Is there someone tell me how to fix it?

[leeivan1007]

@googlemike
it can add the code with:

def callback(lcl, glb):
    # stop training if reward exceeds 199
    total = sum(lcl['episode_rewards'][-101:-1]) / 100
    totalt = lcl['t']
    #print("totalt")
    #print(totalt)
    is_solved = totalt > 2000 and total >= 10
    return is_solved
load_path = 'kuka_model.pkl'
act = deepq.learn(env, network='mlp',lr=1e-3,total_timesteps=0,
buffer_size=50000,print_freq=10,load_path=load_path,callback=callback)

it is not a good solution, but can save this problem.

[ChenyangRan]

@googlemike
it can add the code with:

def callback(lcl, glb):
    # stop training if reward exceeds 199
    total = sum(lcl['episode_rewards'][-101:-1]) / 100
    totalt = lcl['t']
    #print("totalt")
    #print(totalt)
    is_solved = totalt > 2000 and total >= 10
    return is_solved
load_path = 'kuka_model.pkl'
act = deepq.learn(env, network='mlp',lr=1e-3,total_timesteps=0,
buffer_size=50000,print_freq=10,load_path=load_path,callback=callback)

it is not a good solution, but can save this problem.

Hi, I test the code again. It seems that no error about baselines.
Now, it shows that:

No file named 'kuka_model.pkl'.

It means that I should run train_kuka_grasping.py, then run enjoy_kuka_grasping.py?
I don't know well about that.

[leeivan1007]

yes, you are right!
maybe you can modify total_timesteps to 10, make sure this can run first.
but it look like will not be converge #1386
I am trying the solution
maybe we can change the wechat ID,if I have some new progress, I will share it with you.