huangfu/InteractionImitation
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add value dice ray config

Browse Source
This commit is contained in:
Johannes Fischer
2021-08-05 18:33:34 +02:00
parent 8bce5d15f6
commit bf4c19a4d0
2 changed files with 80 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

22
experiments/experiment.py
View File

@@ -90,6 +90,9 @@ def get_full_config(ray_config:dict, method:str)->dict:
if method == 'bc': if method == 'bc':
from src.bc import bc_config from src.bc import bc_config
config = bc_config(ray_config) config = bc_config(ray_config)
elif method == 'vd':
from src.value_dice import vd_config
config = vd_config(ray_config)
else: else:
raise NotImplementedError raise NotImplementedError
return config return config
@@ -118,6 +121,25 @@ def get_ray_config(method:str)->dict:
"head_hidden_dim": tune.lograndint(16,257), "head_hidden_dim": tune.lograndint(16,257),
"head_final_activation": tune.choice(['sigmoid', None]), "head_final_activation": tune.choice(['sigmoid', None]),
} }
elif method == 'vd':
ray_config = {
"policy_lr": tune.loguniform(1e-5, 1e-3),
"value_lr": tune.loguniform(1e-5, 1e-3),
"policy_weight_decay": tune.choice([0, 0.1]),
"value_weight_decay": tune.choice([0, 0.1]),
"train_batch_size": tune.choice([16,32,64]),
"deepsets_phi_hidden_n": tune.randint(1,5),
"deepsets_phi_hidden_dim": tune.lograndint(8,65),
"deepsets_latent_dim": tune.lograndint(8,129),
"deepsets_rho_hidden_n": tune.randint(0,3),
"deepsets_rho_hidden_dim": tune.lograndint(8,129),
"deepsets_output_dim": tune.lograndint(4,129),
"head_hidden_n": tune.randint(1,6),
"head_hidden_dim": tune.lograndint(16,257),
"head_final_activation": tune.choice(['sigmoid', None]),
"clip_grad_norm": tune.choice([.5, 1., 5., 10.]),
"discount": tune.choice([.95, .99])
}
else: else:
raise NotImplementedError raise NotImplementedError
return ray_config return ray_config

83
src/value_dice/value_dice.py
View File

@@ -11,39 +11,72 @@ from src.util.nn_training import optimizer_factory
from tqdm import tqdm from tqdm import tqdm
import json5 import json5
from ray import tune from ray import tune
def vd_config(ray_config): def vd_config(ray_config):
config = { config = {
'ego_encoder': {'input_dim': 5, 'hidden_n': 0, 'hidden_dim':0, 'output_dim': 0}, 'policy_net': {
'deepsets': { 'ego_encoder': {'input_dim': 5, 'hidden_n': 0, 'hidden_dim':0, 'output_dim': 0},
'input_dim': 6, 'deepsets': {
'phi': { 'input_dim': 6,
'hidden_n': ray_config['deepsets_phi_hidden_n'], 'phi': {
'hidden_dim': ray_config['deepsets_phi_hidden_dim'] 'hidden_n': ray_config['deepsets_phi_hidden_n'],
}, 'hidden_dim': ray_config['deepsets_phi_hidden_dim']
'latent_dim': ray_config['deepsets_latent_dim'], },
'rho': { 'latent_dim': ray_config['deepsets_latent_dim'],
'hidden_n': ray_config['deepsets_rho_hidden_n'], 'rho': {
'hidden_dim': ray_config['deepsets_rho_hidden_dim'] 'hidden_n': ray_config['deepsets_rho_hidden_n'],
}, 'hidden_dim': ray_config['deepsets_rho_hidden_dim']
'output_dim': ray_config['deepsets_output_dim'] },
'output_dim': ray_config['deepsets_output_dim']
},
'path_encoder': {'input_dim': 40, 'hidden_n': 0, 'hidden_dim': 0, 'output_dim': 0},
'head': {
'input_dim': 0, # computed in constructor
'hidden_n': ray_config['head_hidden_n'],
'hidden_dim': ray_config['head_hidden_dim'],
'output_dim': 1, # number of outputs e.g. number of actions, or just one
'final_activation': ray_config['head_final_activation'],
},
}, },
'path_encoder': {'input_dim': 40, 'hidden_n': 0, 'hidden_dim': 0, 'output_dim': 0}, 'value_net': {
'head': { 'ego_encoder': {'input_dim': 5, 'hidden_n': 0, 'hidden_dim':0, 'output_dim': 0},
'input_dim': 0, # computed in constructor 'deepsets': {
'hidden_n': ray_config['head_hidden_n'], 'input_dim': 6,
'hidden_dim': ray_config['head_hidden_dim'], 'phi': {
'output_dim': 1, # number of outputs e.g. number of actions, or just one 'hidden_n': ray_config['deepsets_phi_hidden_n'],
'final_activation': ray_config['head_final_activation'], 'hidden_dim': ray_config['deepsets_phi_hidden_dim']
},
'latent_dim': ray_config['deepsets_latent_dim'],
'rho': {
'hidden_n': ray_config['deepsets_rho_hidden_n'],
'hidden_dim': ray_config['deepsets_rho_hidden_dim']
},
'output_dim': ray_config['deepsets_output_dim']
},
'path_encoder': {'input_dim': 40, 'hidden_n': 0, 'hidden_dim': 0, 'output_dim': 0},
'action_dim': 1,
'head': {
'input_dim': 0, # computed in constructor
'hidden_n': ray_config['head_hidden_n'],
'hidden_dim': ray_config['head_hidden_dim'],
'output_dim': 1, # number of outputs e.g. number of actions, or just one
'final_activation': ray_config['head_final_activation'],
},
}, },
'optim': { 'policy_optim': {
'optimizer':'adam', 'optimizer':'adam',
'lr':ray_config['lr'], 'lr':ray_config['policy_lr'],
'weight_decay':ray_config['weight_decay'] 'weight_decay':ray_config['policy_weight_decay']
},
'value_optim': {
'optimizer':'adam',
'lr':ray_config['value_lr'],
'weight_decay':ray_config['value_weight_decay']
}, },
'train_epochs': 40, 'train_epochs': 40,
'train_batch_size': ray_config['train_batch_size'], 'train_batch_size': ray_config['train_batch_size'],
'loss': ray_config['loss'], 'discount': ray_config['discount'],
'clip_grad_norm': ray_config['clip_grad_norm'],
} }
return config return config