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

Merge branch 'main' into idm_upgrade

Browse Source
This commit is contained in:
Johannes Fischer
2022-03-01 09:58:07 +01:00
parent 2965dc9982 57a42f70ec
commit f93e130498
41 changed files with 94 additions and 298 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
bc-experiment.py
View File

@@ -142,7 +142,7 @@ if __name__ == '__main__':
import argparse import argparse
parser = argparse.ArgumentParser() parser = argparse.ArgumentParser()
parser.add_argument('--train', choices=['A', 'B']) parser.add_argument('--train', choices=['A', 'B'])
parser.add_argument('--epochs', type=int, default=1000) parser.add_argument('--epochs', type=int, default=500)
parser.add_argument('--test', type=str, help='path to config file to run final training on') parser.add_argument('--test', type=str, help='path to config file to run final training on')
parser.add_argument('--test_seeds', type=int, default=5) parser.add_argument('--test_seeds', type=int, default=5)
parser.add_argument('--test_cpus', type=int, help='number of cpus available to split test seed training over') parser.add_argument('--test_cpus', type=int, help='number of cpus available to split test seed training over')
@@ -162,10 +162,10 @@ if __name__ == '__main__':
}, },
'policy': { 'policy': {
'learning_rate': 3e-4, 'learning_rate': 3e-4,
'learning_rate_decay': 1.0, 'learning_rate_decay': tune.grid_search([0.999, 1.0]),
'hidden_layer_size': tune.grid_search([20, 40]), 'hidden_layer_size': tune.grid_search([10, 20, 40]),
'n_hidden_layers': tune.grid_search([2, 3]), 'n_hidden_layers': tune.grid_search([2, 3]),
'activation':0, 'activation':tune.grid_search([0, 1]),
}, },
'train_epochs': args.epochs, 'train_epochs': args.epochs,
'seed': 0, 'seed': 0,
@@ -209,4 +209,6 @@ if __name__ == '__main__':
s = analysis._checkpoints[i]['config']['seed'] s = analysis._checkpoints[i]['config']['seed']
check_dir = analysis._checkpoints[i]['logdir'] check_dir = analysis._checkpoints[i]['logdir']
shutil.copyfile(os.path.join(check_dir,'policy_final.pt'), shutil.copyfile(os.path.join(check_dir,'policy_final.pt'),
os.path.join(savepath, f'policy_seed{s}.pt')) os.path.join(savepath, f'policy_seed{s}.pt'))
shutil.copyfile(os.path.join(check_dir,'params.json'),
os.path.join(savepath, 'config.json')) # copy config automatically

33
best_configs/hail-etienne2_expA.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": null,
"abort_unsafe_collision_method": null
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 10,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/hail-etienne2_expB.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": null,
"abort_unsafe_collision_method": null
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 10,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/hail-etienne3_expB.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": null,
"abort_unsafe_collision_method": null
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 40,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/hail-etienne_expA.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": null,
"abort_unsafe_collision_method": null
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 20,
"n_hidden_layers": 4,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/shail-etienne2_expA.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": "circle",
"abort_unsafe_collision_method": "circle"
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 10,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/shail-etienne2_expB.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": "circle",
"abort_unsafe_collision_method": "circle"
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 10,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/shail-etienne3_expB.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": "circle",
"abort_unsafe_collision_method": "circle"
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 40,
"n_hidden_layers": 3,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
best_configs/shail-etienne_expA.json
View File

@@ -1,33 +0,0 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false,
"safe_actions_collision_method": "circle",
"abort_unsafe_collision_method": "circle"
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"clip_ratio": 0.2,
"iterations_per_epoch": 100,
"hidden_layer_size": 20,
"n_hidden_layers": 4,
"activation": 0,
"option": 0
},
"value": {
"learning_rate": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 3,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

25
eval_experiments.py
View File

@@ -1,17 +1,23 @@
import os import os
from src.eval_main import eval_main from src.eval_main import eval_main
from src.evaluation.utils import load_and_average from src.evaluation.utils import load_and_average
import torch
import json
activations = [torch.nn.Tanh, torch.nn.LeakyReLU]
def main(method:str='expert', folder:str=None, locations=[(0,0)], skip_running=False): def main(method:str='expert', folder:str=None, locations=[(0,0)], skip_running=False):
exclude_keys_from_policy_kwargs = {'learning_rate', 'learning_rate_decay', 'clip_ratio', 'iterations_per_epoch', 'option'}
policy_kwargs = {} policy_kwargs = {}
if method in ['expert', 'idm']: if method in ['expert', 'idm']:
env, env_kwargs ='NRasterizedRouteIncrementingAgent', {} env, env_kwargs ='NRasterizedRouteIncrementingAgent', {}
elif method in ['bc','gail']: elif method in ['bc','gail']:
env='NormalizedContinuousEvalEnv' env='NormalizedContinuousEvalEnv'
env_kwargs={'stop_on_collision':True, 'max_episode_steps':1000} env_kwargs={'stop_on_collision':True, 'max_episode_steps':1000}
elif method in ['hail']: elif method in ['hail']:
env = 'NormalizedOptionsEvalEnv' env = 'NormalizedSafeOptionsEvalEnv'
env_kwargs={'stop_on_collision':True, 'max_episode_steps':1000, 'safe_actions_collision_method': None, 'abort_unsafe_collision_method': None} env_kwargs={'stop_on_collision':True, 'max_episode_steps':1000, 'safe_actions_collision_method': None, 'abort_unsafe_collision_method': None}
elif method in ['shail']: elif method in ['shail']:
env = 'NormalizedSafeOptionsEvalEnv' env = 'NormalizedSafeOptionsEvalEnv'
@@ -23,7 +29,18 @@ def main(method:str='expert', folder:str=None, locations=[(0,0)], skip_running=F
if folder is not None: if folder is not None:
files = [os.path.join(folder, f) for f in os.listdir(folder) if os.path.isfile(os.path.join(folder, f))] files = [os.path.join(folder, f) for f in os.listdir(folder) if os.path.isfile(os.path.join(folder, f))]
print('%i folders found in %s folder' %(len(files), folder)) files = [f for f in files if f.endswith('.pt')]
with open(os.path.join(folder, 'config.json'), 'rb') as f:
config = json.load(f)
print('%i policy files found in %s folder' %(len(files), folder))
print('found policy config', config['policy'])
policy_config = {k: v for k, v in config['policy'].items() if k not in exclude_keys_from_policy_kwargs}
policy_config['activation'] = activations[policy_config['activation']]
print('final policy config', policy_config)
policy_kwargs.update(policy_config)
print('final policy kwargs', policy_kwargs)
if not skip_running: if not skip_running:
for policy_file in files: for policy_file in files:
@@ -60,7 +77,7 @@ def latex_print(am, light=False):
print('success rate, distance travelled, RWSE_10, |DeltaV|, AccelJSD') print('success rate, distance travelled, RWSE_10, |DeltaV|, AccelJSD')
if light: if light:
if 'rwse_10s' in am.keys(): if 'rwse_10s' in am.keys():
print("%2.1f& %2.1f & %1.2f & %2.1f& " print("%2.1f& %2.1f & %2.1f & %1.2f& "
"%0.3f \\\\" %( 100*am['success rate'][0], am['mean travel distance'][0], am['rwse_10s'][0], "%0.3f \\\\" %( 100*am['success rate'][0], am['mean travel distance'][0], am['rwse_10s'][0],
am['average absolute average velocity'][0],am['acceleration distribution divergence'][0] )) am['average absolute average velocity'][0],am['acceleration distribution divergence'][0] ))
return return
@@ -71,7 +88,7 @@ def latex_print(am, light=False):
return return
print("%2.1f \\scriptstyle\\pm %2.1f & %2.1f \\scriptstyle\\pm %2.1f & " print("%2.1f \\scriptstyle\\pm %2.1f & %2.1f \\scriptstyle\\pm %2.1f & "
"%1.2f \\scriptstyle\\pm %1.2f & %2.1f \\scriptstyle\\pm %1.1f & " "%2.1f \\scriptstyle\\pm %1.1f & %1.2f \\scriptstyle\\pm %1.2f & "
"%0.3f \\scriptstyle\\pm %0.3f \\\\" %( 100*am['success rate'][0], 100*am['success rate'][1], "%0.3f \\scriptstyle\\pm %0.3f \\\\" %( 100*am['success rate'][0], 100*am['success rate'][1],
am['mean travel distance'][0] , am['mean travel distance'][1] , am['mean travel distance'][0] , am['mean travel distance'][1] ,
am['rwse_10s'][0] , am['rwse_10s'][1] , am['rwse_10s'][0] , am['rwse_10s'][1] ,

4
gail-experiment.py
View File

@@ -235,4 +235,6 @@ if __name__ == '__main__':
s = analysis._checkpoints[i]['config']['seed'] s = analysis._checkpoints[i]['config']['seed']
check_dir = analysis._checkpoints[i]['logdir'] check_dir = analysis._checkpoints[i]['logdir']
shutil.copyfile(os.path.join(check_dir,'policy_final.pt'), shutil.copyfile(os.path.join(check_dir,'policy_final.pt'),
os.path.join(savepath, f'policy_seed{s}.pt')) os.path.join(savepath, f'policy_seed{s}.pt'))
shutil.copyfile(os.path.join(check_dir,'params.json'),
os.path.join(savepath, 'config.json')) # copy config automatically

BIN
out/hail/expA/loc_r0t0/policy_seed1_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed1_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed2_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed2_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed3_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed3_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed4_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed4_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed5_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expA/loc_r0t0/policy_seed5_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed1_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed1_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed2_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed2_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed3_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed3_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed4_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed4_tseed0_summary.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed5_tseed0_comparison.pkl Normal file
View File

Binary file not shown.

BIN
out/hail/expB/loc_r0t4/policy_seed5_tseed0_summary.pkl Normal file
View File

Binary file not shown.

11
shail-experiment.py
View File

@@ -247,8 +247,17 @@ if __name__ == '__main__':
os.makedirs(savepath) os.makedirs(savepath)
import shutil import shutil
# save config
shutil.copyfile(
args.test,
os.path.join(savepath, 'config.json')
)
for i in range(args.test_seeds): for i in range(args.test_seeds):
s = analysis._checkpoints[i]['config']['seed'] s = analysis._checkpoints[i]['config']['seed']
check_dir = analysis._checkpoints[i]['logdir'] check_dir = analysis._checkpoints[i]['logdir']
shutil.copyfile(os.path.join(check_dir,'policy_final.pt'), shutil.copyfile(os.path.join(check_dir,'policy_final.pt'),
os.path.join(savepath, f'policy_seed{s}.pt')) os.path.join(savepath, f'policy_seed{s}.pt'))
shutil.copyfile(os.path.join(check_dir,'params.json'),
os.path.join(savepath, 'config.json')) # copy config automatically

14
src/eval_main.py
View File

@@ -41,33 +41,33 @@ def load_policy(method:str,
if method == 'idm': if method == 'idm':
policy = IDMRulePolicy(env, **policy_kwargs) policy = IDMRulePolicy(env, **policy_kwargs)
elif method == 'bc': elif method == 'bc':
policy = SetPolicy(env.action_space.shape[-1]) policy = SetPolicy(env.action_space.shape[-1], **policy_kwargs)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'gail-trpo': elif method == 'gail-trpo':
policy = SetPolicy(env.action_space.shape[-1]) policy = SetPolicy(env.action_space.shape[-1], **policy_kwargs)
policy(torch.zeros(env.observation_space.shape)) policy(torch.zeros(env.observation_space.shape))
policy = ReparamPolicy(policy) policy = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'gail': elif method == 'gail':
policy = SetPolicy(env.action_space.shape[-1]) policy = SetPolicy(env.action_space.shape[-1], **policy_kwargs)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'rail': elif method == 'rail':
raise NotImplementedError raise NotImplementedError
elif method == 'hail-trpo': elif method == 'hail-trpo':
policy = SetDiscretePolicy(env.action_space.n) policy = SetMaskedDiscretePolicy(env.action_space.n, **policy_kwargs)
policy(torch.zeros(env.observation_space.shape)) policy(torch.zeros(env.observation_space.shape))
policy = ReparamPolicy(policy) policy = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'hail': elif method == 'hail':
policy = SetDiscretePolicy(env.action_space.n) policy = SetMaskedDiscretePolicy(env.action_space.n, **policy_kwargs)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'shail-trpo': elif method == 'shail-trpo':
policy = SetMaskedDiscretePolicy(env.action_space.n) policy = SetMaskedDiscretePolicy(env.action_space.n, **policy_kwargs)
policy( policy(
torch.zeros(env.observation_space['observation'].shape), torch.zeros(env.observation_space['observation'].shape),
torch.zeros(env.observation_space['safe_actions'].shape) torch.zeros(env.observation_space['safe_actions'].shape)
@@ -76,7 +76,7 @@ def load_policy(method:str,
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
elif method == 'shail': elif method == 'shail':
policy = SetMaskedDiscretePolicy(env.action_space.n) policy = SetMaskedDiscretePolicy(env.action_space.n, **policy_kwargs)
policy.load_state_dict(torch.load(policy_file, map_location=ml)) policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval() policy.eval()
else: else:

15
test_policies/bc/expA/config.json Normal file
View File

@@ -0,0 +1,15 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"hidden_layer_size": 40,
"n_hidden_layers": 2,
"activation": 0
},
"train_epochs": 300,
"seed": 0
}

15
test_policies/bc/expB/config.json Normal file
View File

@@ -0,0 +1,15 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false
},
"policy": {
"learning_rate": 0.0003,
"learning_rate_decay": 1.0,
"hidden_layer_size": 40,
"n_hidden_layers": 2,
"activation": 0
},
"train_epochs": 300,
"seed": 0
}

2
best_configs/gail-trpo_expA.json → test_policies/gail/expA/config.json
View File

@@ -6,7 +6,7 @@
"policy": { "policy": {
"learning_rate": 0.0003, "learning_rate": 0.0003,
"learning_rate_decay": 1.0, "learning_rate_decay": 1.0,
"delta": 0.01, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 40, "hidden_layer_size": 40,
"n_hidden_layers": 2, "n_hidden_layers": 2,

2
best_configs/gail-trpo_expB.json → test_policies/gail/expB/config.json
View File

@@ -6,7 +6,7 @@
"policy": { "policy": {
"learning_rate": 0.0003, "learning_rate": 0.0003,
"learning_rate_decay": 1.0, "learning_rate_decay": 1.0,
"delta": 0.01, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 40, "hidden_layer_size": 40,
"n_hidden_layers": 2, "n_hidden_layers": 2,

8
best_configs/hail-etienne3_expA.json → test_policies/hail/expA/config.json
View File

@@ -11,7 +11,7 @@
"clip_ratio": 0.2, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 40, "hidden_layer_size": 40,
"n_hidden_layers": 3, "n_hidden_layers": 2,
"activation": 0, "activation": 0,
"option": 0 "option": 0
}, },
@@ -23,11 +23,11 @@
"learning_rate": 0.001, "learning_rate": 0.001,
"weight_decay": 0.0001, "weight_decay": 0.0001,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"n_hidden_layers_element": 3, "n_hidden_layers_element": 4,
"n_hidden_layers_global": 2, "n_hidden_layers_global": 1,
"hidden_layer_size": 10, "hidden_layer_size": 10,
"activation": 0 "activation": 0
}, },
"train_epochs": 100, "train_epochs": 90,
"seed": 0 "seed": 0
} }

6
best_configs/hail-etienne_expB.json → test_policies/hail/expB/config.json
View File

@@ -11,7 +11,7 @@
"clip_ratio": 0.2, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 20, "hidden_layer_size": 20,
"n_hidden_layers": 4, "n_hidden_layers": 2,
"activation": 0, "activation": 0,
"option": 0 "option": 0
}, },
@@ -23,11 +23,11 @@
"learning_rate": 0.001, "learning_rate": 0.001,
"weight_decay": 0.0001, "weight_decay": 0.0001,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"n_hidden_layers_element": 3, "n_hidden_layers_element": 4,
"n_hidden_layers_global": 2, "n_hidden_layers_global": 2,
"hidden_layer_size": 10, "hidden_layer_size": 10,
"activation": 0 "activation": 0
}, },
"train_epochs": 100, "train_epochs": 85,
"seed": 0 "seed": 0
} }

8
best_configs/shail-etienne3_expA.json → test_policies/shail/expA/config.json
View File

@@ -11,7 +11,7 @@
"clip_ratio": 0.2, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 40, "hidden_layer_size": 40,
"n_hidden_layers": 3, "n_hidden_layers": 2,
"activation": 0, "activation": 0,
"option": 0 "option": 0
}, },
@@ -23,11 +23,11 @@
"learning_rate": 0.001, "learning_rate": 0.001,
"weight_decay": 0.0001, "weight_decay": 0.0001,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"n_hidden_layers_element": 3, "n_hidden_layers_element": 4,
"n_hidden_layers_global": 2, "n_hidden_layers_global": 1,
"hidden_layer_size": 10, "hidden_layer_size": 10,
"activation": 0 "activation": 0
}, },
"train_epochs": 100, "train_epochs": 90,
"seed": 0 "seed": 0
} }

6
best_configs/shail-etienne_expB.json → test_policies/shail/expB/config.json
View File

@@ -11,7 +11,7 @@
"clip_ratio": 0.2, "clip_ratio": 0.2,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"hidden_layer_size": 20, "hidden_layer_size": 20,
"n_hidden_layers": 4, "n_hidden_layers": 2,
"activation": 0, "activation": 0,
"option": 0 "option": 0
}, },
@@ -23,11 +23,11 @@
"learning_rate": 0.001, "learning_rate": 0.001,
"weight_decay": 0.0001, "weight_decay": 0.0001,
"iterations_per_epoch": 100, "iterations_per_epoch": 100,
"n_hidden_layers_element": 3, "n_hidden_layers_element": 4,
"n_hidden_layers_global": 2, "n_hidden_layers_global": 2,
"hidden_layer_size": 10, "hidden_layer_size": 10,
"activation": 0 "activation": 0
}, },
"train_epochs": 100, "train_epochs": 85,
"seed": 0 "seed": 0
} }