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

Merge branch 'main' of https://github.com/sisl/InteractionImitation into main

Browse Source
This commit is contained in:
Arec Jamgochian
2022-02-28 03:35:03 -08:00
parent 5ada1cc543 a9314c4657
commit e602aa0641
15 changed files with 389 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
eval_experiments.py
View File

@@ -1,17 +1,23 @@
import os
from src.eval_main import eval_main
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):
exclude_keys_from_policy_kwargs = {'learning_rate', 'learning_rate_decay', 'clip_ratio', 'iterations_per_epoch', 'option'}
policy_kwargs = {}
if method in ['expert', 'idm']:
env, env_kwargs ='NRasterizedRouteIncrementingAgent', {}
elif method in ['bc','gail']:
env='NormalizedContinuousEvalEnv'
env_kwargs={'stop_on_collision':True, 'max_episode_steps':1000}
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}
elif method in ['shail']:
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:
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:
for policy_file in files:

7
shail-experiment.py
View File

@@ -247,6 +247,13 @@ if __name__ == '__main__':
os.makedirs(savepath)
import shutil
# save config
shutil.copyfile(
args.test,
os.path.join(savepath, 'config.json')
)
for i in range(args.test_seeds):
s = analysis._checkpoints[i]['config']['seed']
check_dir = analysis._checkpoints[i]['logdir']

14
src/eval_main.py
View File

@@ -41,33 +41,33 @@ def load_policy(method:str,
if method == 'idm':
policy = IDMRulePolicy(env, **policy_kwargs)
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.eval()
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 = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
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.eval()
elif method == 'rail':
raise NotImplementedError
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 = ReparamPolicy(policy)
policy.load_state_dict(torch.load(policy_file, map_location=ml))
policy.eval()
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.eval()
elif method == 'shail-trpo':
policy = SetMaskedDiscretePolicy(env.action_space.n)
policy = SetMaskedDiscretePolicy(env.action_space.n, **policy_kwargs)
policy(
torch.zeros(env.observation_space['observation'].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.eval()
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.eval()
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
}

31
test_policies/gail/expA/config.json Normal file
View File

@@ -0,0 +1,31 @@
{
"experiment": "A",
"trainenv": {
"stop_on_collision": false
},
"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": 2,
"activation": 0
},
"value": {
"learning_rate": 0.0001,
"weight_decay": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 4,
"n_hidden_layers_global": 1,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

31
test_policies/gail/expB/config.json Normal file
View File

@@ -0,0 +1,31 @@
{
"experiment": "B",
"trainenv": {
"stop_on_collision": false
},
"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": 2,
"activation": 0
},
"value": {
"learning_rate": 0.0001,
"weight_decay": 0.001,
"iterations_per_epoch": 1000
},
"discriminator": {
"learning_rate": 0.001,
"weight_decay": 0.0001,
"iterations_per_epoch": 100,
"n_hidden_layers_element": 4,
"n_hidden_layers_global": 1,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 100,
"seed": 0
}

33
test_policies/hail-etienne/expA/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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
test_policies/hail-etienne/expB/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 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
test_policies/hail/expA/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 40,
"n_hidden_layers": 2,
"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": 4,
"n_hidden_layers_global": 1,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 90,
"seed": 0
}

33
test_policies/hail/expB/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 20,
"n_hidden_layers": 2,
"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": 4,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 85,
"seed": 0
}

33
test_policies/shail-etienne/expA/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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
}

33
test_policies/shail-etienne/expB/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 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
test_policies/shail/expA/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 40,
"n_hidden_layers": 2,
"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": 4,
"n_hidden_layers_global": 1,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 90,
"seed": 0
}

33
test_policies/shail/expB/config.json Normal file
View File

@@ -0,0 +1,33 @@
{
"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": 20,
"n_hidden_layers": 2,
"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": 4,
"n_hidden_layers_global": 2,
"hidden_layer_size": 10,
"activation": 0
},
"train_epochs": 85,
"seed": 0
}