def main():
parser = arg_parser()
parser.add_argument('--env',
help='environment ID',
type=str,
default='Swimmer-v2')
parser.add_argument('--dir', type=str, default='EXP_ON_fix_norm')
parser.add_argument('--thesis', type=str, default='Online_V0')
args = parser.parse_args()
# dirname = '~/Desktop/carla_sample_efficient/data/bk/bkup_EXP1_FINAL/'+args.extra_dir+args.env
dirname = '~/Desktop/ppo_test/' + args.dir + '/' + args.env
results = pu.load_results(dirname)
# r_copos1,r_copos2,r_trpo,r_ppo=filt(results,'copos1'),filt(results,'copos2'),filt(results,'trpo'),filt(results,'ppo')
# r_sil_n2=filt(results,'sil_n2_l0.001')
# dt={'copos1':r_copos1, 'copos2':r_copos2,'trpo':r_trpo, 'ppo':r_ppo, 'sil_slight':r_sil_n2}
r_ppo = filt(results, 'ppo')
dt = {'ppo': r_ppo}
for name in dt:
pu.plot_results(dt[name],
xy_fn=pu.progress_default_xy_fn,
average_group=True,
split_fn=lambda _: '',
shaded_err=True,
shaded_std=False)
plt.xlabel('Number of Timesteps [M]')
plt.ylabel('Average Return [-]')
plt.tight_layout()
fig = plt.gcf()
fig.set_size_inches(9, 7.5)
# fig.savefig("/Users/zsbjltwjj/Desktop/carla_sample_efficient/plot_f/ONLINE/"+args.extra_dir+args.env+'/'+name+'.pdf', format='pdf')
# fig.savefig("/Users/zsbjltwjj/Desktop/thesis/img/"+args.thesis+"/"+args.env+'/'+name+'.pdf', format="pdf")
fig.savefig("/Users/zsbjltwjj/Desktop/ppo_test/" + args.dir + '-' +
name + '.pdf',
format="pdf")
def main():
parser = arg_parser()
parser.add_argument('--env',
help='environment ID',
type=str,
default='HalfCheetah-v2')
parser.add_argument('--st_seed',
help='start number of seeds',
type=int,
default=0)
parser.add_argument('--seeds', help='number of seeds', type=int, default=1)
parser.add_argument('--num_timesteps', type=str, default="3e4")
parser.add_argument('--filename',
type=str,
default='_Offline_Evaluation_nosil.png')
args = parser.parse_args()
if args.env == 'Swimmer-v2' or args.env == 'HalfCheetah-v2':
mbl_args = '--num_samples=1500 --num_elites=10 --horizon=10 --eval_freq=10 --mbl_train_freq=10'
elif arg.env == 'Reacher-v2' or args.env == 'Ant-v2':
mbl_args = '--num_samples=1500 --num_elites=10 --horizon=5 --eval_freq=10 --mbl_train_freq=10'
# algo_names=["ppo2_sil_online","copos_sil_online","ppo2_online","copos_online"]
# legend_names=["ppo2+sil","copos+sil","ppo2","copos"]
# argus=["","","",""]
algo_names = [
"mbl_ppo2", "ppo2_offline", "mbl_copos", "copos_offline", "mbl_trpo",
"trpo_offline"
]
# algo_names=["mbl_ppo2","ppo2_offline",
# "mbl_copos","copos_offline"]
legend_names = [
"mbl+ppo2", "ppo2", "mbl+copos", "copos", "mbl+trpo", "trpo"
]
# legend_names=["mbl+ppo2","ppo2",
# "mbl+copos","copos"]
#argus=['--num_samples=1 --num_elites=1 --horizon=2' for _ in range(len(algo_names))]
argus = [mbl_args for _ in range(len(algo_names))]
for i in range(args.st_seed, args.st_seed + args.seeds):
for j in range(len(algo_names)):
os.system("python ../algos/" + algo_names[j] + "/run.py --alg=" +
algo_names[j] + " --num_timestep=" + args.num_timesteps +
" --seed=" + str(i) + " --env=" + args.env +
" --log_path=~/Desktop/logs/EXP2_nosil/" + args.env +
"/" + legend_names[j] + "-" + str(i) + ' ' + argus[j])
results = pu.load_results('~/Desktop/logs/EXP2_nosil/' + args.env)
pu.plot_results(results,
xy_fn=pu.progress_itermbl_xy_fn,
average_group=True,
split_fn=lambda _: '')
#plt.title(args.env+" Online Evaluation")
plt.xlabel('Evaluation Epochs [-]')
plt.ylabel('Average Return [-]')
fig = plt.gcf()
fig.set_size_inches(9.5, 7.5)
fig.savefig(args.env + "_" + args.filename)
def plot_results(dirs, num_timesteps=10e6, xaxis=X_TIMESTEPS, yaxis=Y_REWARD, title='', split_fn=split_by_task, resample=100):
results = plot_util.load_results(dirs)
new_results = []
for result in results:
dfs = result.monitor
for df in dfs:
temp_result = copy.deepcopy(result)
temp_result = temp_result._replace(monitor=df)
new_results.append(temp_result)
plot_util.plot_results(new_results, xy_fn=lambda r: ts2xy(r.monitor, xaxis, yaxis), group_fn=split_fn, average_group=True, resample=resample)
def main():
parser = arg_parser()
parser.add_argument('--env',
help='environment ID',
type=str,
default='HalfCheetah-v2')
parser.add_argument('--extra_dir', type=str, default='')
args = parser.parse_args()
dirname = '~/Desktop/carla_sample_efficient/data/bk/bkup_EXP1_FINAL/' + args.extra_dir + args.env
results = pu.load_results(dirname)
r_copos, r_trpo, r_ppo = filt(results,
'copos'), filt(results,
'trpo'), filt(results, 'ppo')
r_sil_n2 = filt(results, 'sil_n2_l0.001')
dt = {
'copos': r_copos,
'trpo': r_trpo,
'ppo': r_ppo,
'sil_slight': r_sil_n2
}
for name in dt:
pu.plot_results(dt[name],
xy_fn=pu.progress_default_xy_fn,
average_group=True,
split_fn=lambda _: '',
shaded_err=True,
shaded_std=False)
plt.xlabel('Number of Timesteps [M]')
plt.ylabel('Average Return [-]')
plt.tight_layout()
fig = plt.gcf()
fig.set_size_inches(9, 7.5)
fig.savefig(
"/Users/zsbjltwjj/Desktop/carla_sample_efficient/plot_f/ONLINE/" +
args.env + '/' + name + '.pdf',
format='pdf')
if name == 'sil_slight':
pu.plot_results(dt[name],
xy_fn=pu.progress_default_entropy_xy_fn,
average_group=True,
split_fn=lambda _: '',
shaded_err=True,
shaded_std=False,
legend_entropy=1)
plt.xlabel('Number of Timesteps [M]')
plt.ylabel('Entropy [-]')
plt.tight_layout()
fig = plt.gcf()
fig.set_size_inches(9, 7.5)
fig.savefig(
"/Users/zsbjltwjj/Desktop/carla_sample_efficient/plot_f/ONLINE/"
+ args.env + '/' + name + '_entropy.pdf',
format="pdf")
def plot_results(dirs,
num_timesteps=10e6,
xaxis=X_TIMESTEPS,
yaxis=Y_REWARD,
title='',
split_fn=split_by_task):
results = plot_util.load_results(dirs)
plot_util.plot_results(results,
split_fn=split_fn,
average_group=True,
resample=int(1e6))
def test_plot_util():
nruns = 4
logdirs = [smoketest('--alg=ppo2 --env=CartPole-v0 --num_timesteps=10000') for _ in range(nruns)]
data = pu.load_results(logdirs)
assert len(data) == 4
_, axes = pu.plot_results(data[:1]); assert len(axes) == 1
_, axes = pu.plot_results(data, tiling='vertical'); assert axes.shape==(4,1)
_, axes = pu.plot_results(data, tiling='horizontal'); assert axes.shape==(1,4)
_, axes = pu.plot_results(data, tiling='symmetric'); assert axes.shape==(2,2)
_, axes = pu.plot_results(data, split_fn=lambda _: ''); assert len(axes) == 1
def main():
# results = pu.load_results('data_the_best')
results = pu.load_results('data_Test_obstacle_origin/log_data')
r = results[0]
# plt.plot(np.cumsum(r.monitor.l), r.monitor.r)
# plt.plot(np.cumsum(r.monitor.l), pu.smooth(r.monitor.r, radius=10))
#### plt.plot(r.progress.total_timesteps, r.progress.eprewmean)
# print('keys:', r.progress.keys())
# plt.plot(r.progress['epoch'], r.progress['test/success_rate'])
# plt.plot(r.progress['epoch'], pu.smooth(r.progress['test/success_rate'], radius=5))
# pu.plot_results(results)
pu.plot_results(results, average_group=True, split_fn=lambda _: '')
set_trace()
def plot_accuracy(root_dir):
all_results = load_results(root_dir, verbose=True)
def xy_fn(r):
x = np.cumsum(r.progress['xs'])
y = r.progress['errors']
return x, y
def split_fn(r):
name = r.name
splits = name.split('-')
return splits[0]
def group_fn(r):
name = r.name
splits = name.split('-')
alg_name = splits[1]
if alg_name == 'SGES':
return 'SGES'
elif alg_name == 'CMA':
return 'CMA-ES'
elif alg_name == 'GES':
return 'Guided ES'
elif alg_name == 'ES':
return 'Vanilla ES'
elif alg_name == 'ASEBO':
return 'ASEBO'
else:
raise ValueError('%s not supported' % alg_name)
_all_results = []
for result in all_results:
if 'Sphere' in result.name:
if 'CMA' in result.name:
continue
_all_results.append(result)
all_results = _all_results
plt.figure(dpi=300)
fig, axarr = pu.plot_results(all_results,
xy_fn=xy_fn,
split_fn=split_fn,
group_fn=group_fn,
shaded_std=True,
shaded_err=False,
average_group=True,
tiling='horizontal',
xlabel='Evaluations',
ylabel='Cosine Similarity')
plt.subplots_adjust(hspace=0.2,
wspace=0.2,
bottom=0.2,
left=0.08,
top=0.95)
for ax in axarr[0]:
ax.set_xticks(np.arange(0, 12.5e4, 2.5e4))
ax.set_xticklabels(['0', '25k', '50k', '75k', '100k'])
plt.savefig('blackbox_accuracy.pdf', bbox_inches='tight')
def plot_data(exp, savefig, ttype):
savefig = abspath(savefig)
try:
results = organize_results(pu.load_results(exp))
pu.plot_results(results,
average_group=True,
split_fn=lambda _: '',
xy_fn=ep_distance_ratio_train \
if ttype == 'train' else ep_distance_ratio_test,
shaded_std=False,
shaded_err=True)
if os.path.isfile(savefig): os.remove(savefig)
plt.savefig(savefig)
plt.clf()
#print("Plot saved to: {}".format(savefig))
except Exception as e:
print("Plotting failed for {}".format(savefig))
print("Reason: {}".format(str(e)))
def plot_save_results(xy_fn, file="logs/time_rewards.png"):
f, ax = pu.plot_results(
results,
xy_fn=xy_fn,
split_fn=lambda _: "",
average_group=True,
shaded_err=False,
)
f.savefig(file)
def plot_results(self):
# Create plot directory
os.makedirs(self.plot_dir, exist_ok=True)
results = pu.load_results(
os.path.join(self.log_dir,
self.env_name.split('-')[0], ''))
pu.plot_results(results,
average_group=True,
split_fn=lambda _: '',
shaded_std=False)
plt.xlabel('Timestep')
plt.ylabel('Reward')
fig = plt.gcf()
plot_path = os.path.join(self.plot_dir, 'plot_' + self.env_name)
fig.savefig(plot_path, bbox_inches='tight')
plt.show()
def main():
parser = arg_parser()
parser.add_argument('--env', help='environment ID', type=str, default='HalfCheetah-v2')
parser.add_argument('--extra_dir', type=str, default='')
args = parser.parse_args()
args.extra_dir='EXP2_IAS_5M_TRPO_OFF/'
dirname = '~/Desktop/carla_sample_efficient/data/bk/bkup_EXP2_FINAL/'+args.extra_dir+args.env
results = pu.load_results(dirname)
r_copos_nosil,r_trpo_nosil,r_ppo_nosil=filt(results,'copos-'),filt(results,'trpo-'),filt(results,'ppo-')
r_copos_sil,r_trpo_sil,r_ppo_sil=filt(results,'copos+sil-'),filt(results,'trpo+sil-'),filt(results,'ppo+sil-')
r_mbl_sil=filt(results,'mbl+','sil-')
# r_mbl_nosil_tmp=[r for r in results if r not in r_mbl_sil]
r_mbl_nosil=filt_or_or(results,'mbl+copos-','mbl+trpo-','mbl+ppo-')
r_copos_comp, r_trpo_comp, r_ppo_comp=filt_or(results,'mbl+copos','copos+sil'),filt_or(results,'mbl+trpo','trpo+sil'),filt_or(results,'mbl+ppo','ppo+sil')
# dt={'copos_nosil':r_copos_nosil, 'trpo_nosil':r_trpo_nosil, 'ppo_nosil':r_ppo_nosil,
# 'copos_sil':r_copos_sil, 'trpo_sil':r_trpo_sil, 'ppo_sil':r_ppo_sil,
# 'mbl_nosil':r_mbl_nosil, 'mbl_sil':r_mbl_sil,
# 'copos_comp':r_copos_comp, 'trpo_comp':r_trpo_comp, 'ppo_comp':r_ppo_comp}
r_trpo_all=filt(results,'trpo')
dt={'trpo_research':r_trpo_all}
for name in dt:
pu.plot_results(dt[name],xy_fn=pu.progress_mbl_vbest_xy_fn,average_group=True,name=name,split_fn=lambda _: '',shaded_err=True,shaded_std=False)
plt.xlabel('Number of Timesteps [M]')
plt.ylabel('Best Average Return [-]')
plt.tight_layout()
fig = plt.gcf()
fig.set_size_inches(9, 7.5)
fig.savefig("/Users/zsbjltwjj/Desktop/carla_sample_efficient/plot_f/OFFLINE/"+args.env+'/'+name+'.pdf',format="pdf")
if name=='mbl_nosil' or name=='mbl_sil':
pu.plot_results(dt[name],xy_fn=pu.progress_default_entropy_xy_fn,average_group=True,name=name,split_fn=lambda _: '',shaded_err=True,shaded_std=False,legend_entropy=1)
plt.xlabel('Number of Timesteps [M]')
plt.ylabel('Entropy [-]')
plt.tight_layout()
fig = plt.gcf()
fig.set_size_inches(9, 7.5)
fig.savefig("/Users/zsbjltwjj/Desktop/carla_sample_efficient/plot_f/OFFLINE/"+args.env+'/'+name+'_entropy.pdf',format="pdf")
def plot_k(root_dir):
all_results = load_results(root_dir, verbose=True)
def xy_fn(r):
x = np.cumsum(r.progress['xs'])
y = r.progress['ys']
return x, y
def split_fn(r):
name = r.name
splits = name.split('-')
return splits[0]
def group_fn(r):
name = r.name
splits = name.split('-')
alg_name = splits[1]
if alg_name == 'ES':
return 'Vanilla ES'
elif alg_name == 'SGES100':
return 'SGES(k=100)'
elif alg_name == 'SGES1':
return 'SGES(k=1)'
elif alg_name == 'SGES5':
return 'SGES(k=5)'
else:
return 'SGES(k=%s)' % alg_name[-2:]
plt.figure(dpi=300)
fig, axarr = pu.plot_results(all_results,
xy_fn=xy_fn,
split_fn=split_fn,
group_fn=group_fn,
shaded_std=True,
shaded_err=False,
average_group=True,
tiling='horizontal',
xlabel='Evaluations',
ylabel='Loss')
plt.subplots_adjust(hspace=0.2,
wspace=0.2,
bottom=0.2,
left=0.08,
top=0.95)
for ax in axarr[0]:
ax.set_xticks(np.arange(0, 12.5e4, 2.5e4))
ax.set_xticklabels(['0', '25k', '50k', '75k', '100k'])
# fig.text(0.5, 0.05, s='# Evaluation', fontsize=18)
# fig.text(0.04, 0.5, s='Loss', fontsize=18, rotation='vertical')
plt.savefig('blackbox_k.pdf', bbox_inches='tight')
def plot_monitors():
# If you want to average results for multiple seeds, LOG_DIRS must contain subfolders in the
# following format: <name_exp0>-0, <name_exp0>-1, <name_exp1>-0, <name_exp1>-1.
# Where names correspond to experiments you want to compare separated with random seeds by dash.
LOG_DIRS = '/home/deep3/logs/Humanoid-v2/'
# LOG_DIRS = '/home/deep3/logs/Hopper-v2/'
# Uncomment below to see the effect of the timit limits flag
# LOG_DIRS = 'time_limit_logs/reacher'
results = pu.load_results(LOG_DIRS, running_agents=3)
fig, ax = pu.plot_results(results,
average_group=True,
split_fn=lambda _: '',
shaded_std=False)
from baselines.common import plot_util as pu import matplotlib.pyplot as plt import numpy as np dataDir = 'baselinesData' envName = 'HalfCheetah-v2' alg = 'ppo2' results = pu.load_results(dataDir + '/' + envName) #r = results[0] #plt.plot(np.cumsum(r.monitor.l), r.monitor.r) pu.plot_results(results) plt.show()
def plot_results(dirs, num_timesteps=10e6, xaxis=X_TIMESTEPS, yaxis=Y_REWARD, title='', split_fn=split_by_task):
results = plot_util.load_results(dirs)
plot_util.plot_results(results, xy_fn=lambda r: ts2xy(r['monitor'], xaxis, yaxis), split_fn=split_fn, average_group=True, resample=int(1e6))
from baselines.common import plot_util as pu
import matplotlib.pyplot as plt
results = pu.load_results('path/to/your/experiment01')
pu.plot_results(results,
average_group=True,
split_fn=lambda _: '',
shaded_std=False,
shaded_err=True,
figsize=(10, 6),
smooth_step=10.0)
plt.title('Experiment01', fontsize=30)
plt.tight_layout()
plt.show()
from baselines.common import plot_util as pu
import matplotlib.pyplot as plt
import argparse
import os
"""Create a plot of training data for a given environment using baselines plotting utility
Command line arguments:
--env: environment name (ex: --env=RoboschoolHalfCheetah-v1)
--dir: directory where training data is logged default is ./data/
"""
if __name__ == "__main__":
# Parse command line arguments
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument('--env', help='Environment name')
parser.add_argument('--dir', help='Data directory', default='./data')
args = parser.parse_args()
envName = args.env
# Load results with baselines plot utility
results = pu.load_results('./'+args.dir+'/'+envName)
# Plot results
pu.plot_results(results, average_group=True, split_fn=lambda _: '', shaded_std=False)
# Save plot as a pdf in a subdirectory of the data directory called plots
if not os.path.exists('./'+args.dir+'/plots'):
os.mkdir('./'+args.dir+'/plots')
plt.savefig('./'+args.dir+'/plots/'+envName+'.pdf')
def main():
"""
Plot the plots inside the folder given
"""
# Now plot the common things
args.files = sorted(args.files)
splits = args.files[0].split('/')
if splits[-1] == '':
splits = splits[-3]
else:
splits = splits[-2]
env = splits
results = []
for file in args.files:
print(file)
results.extend(
pu.load_results(file, success=args.success, length=args.length))
# Print details
if args.print:
allrecords = dict()
for i in range(len(results)):
key = check_last_name(results[i])
data = np.array(results[i].monitor)[-10:, 1]
allrecords[key] = allrecords.get(key, []) + [data]
# Print results
for k, v in allrecords.items():
v = np.concatenate(v)
vm = v.mean()
vs = v.std()
print('{} {} {}'.format(k, vm, vs))
return None
fig = pu.plot_results(results,
average_group=True,
shaded_err=False,
shaded_std=True,
max_step=args.max_step,
smooth_step=args.smooth_step,
group_fn=lambda _: check_last_name(_),
split_fn=lambda _: '',
figsize=(10, 10))
# Add results for behaviour cloning if present
'''
allfiles = []
for file in args.files:
for r, dirs, files in os.walk(file):
txtfiles = list(filter(lambda x: x.endswith('bc.txt'), files))
allfiles.extend(list(map(lambda x: os.path.join(r, x), txtfiles)))
if allfiles != []:
bcreward = []
for file in allfiles:
with open(file, 'r') as fi:
meanrew = float(fi.readlines()[0])
bcreward.append(meanrew)
# Get mean and std
mean = np.mean(bcreward)
std = np.std(bcreward)
idxcolor=4
plt.plot([0, args.max_step], [mean, mean], label='BC', color=COLORS[idxcolor])
plt.fill_between([0, args.max_step], [mean - std, mean - std], [mean + std, mean + std], alpha=0.2, color=COLORS[idxcolor])
'''
plt.xlabel('Number of steps', fontsize=20)
plt.ylabel('Reward' if not args.length else 'Episode length', fontsize=20)
#plt.yscale('log')
plt.title(env, fontsize=24)
if args.legend != []:
'''
if allfiles != []:
args.legend.append('BC')
'''
#plt.legend(args.legend, loc='lower right')
plt.legend(args.legend)
#plt.ticklabel_format(useOffset=1)
plt.savefig(
'{}.png'.format(env),
bbox_inches='tight',
)
print("saved ", env)
def main():
"""
Plot the plots inside the folder given
"""
# Now plot the common things
splits = args.files[0].split('/')
if splits[-1] == '':
splits = splits[-2]
else:
splits = splits[-1]
env = splits
results = pu.load_results(args.files, )
fig = pu.plot_results(results,
average_group=True,
shaded_err=False,
shaded_std=True,
group_fn=lambda _: check_last_name(_),
split_fn=lambda _: '',
figsize=(10, 10))
# Add results for behaviour cloning if present
allbcfiles = [args.bcpath]
allfiles = []
allrandomfiles = [] # For random agent behavior
for file in allbcfiles:
for r, dirs, files in os.walk(file):
print(files)
txtfiles = list(filter(lambda x: 'BC_' in x and '.txt' in x,
files))
rndfiles = list(
filter(lambda x: 'random_' in x and '.txt' in x, files))
allfiles.extend(list(map(lambda x: os.path.join(r, x), txtfiles)))
allrandomfiles.extend(
list(map(lambda x: os.path.join(r, x), rndfiles)))
## Show all files for BC and plot
print(allfiles)
if allfiles != []:
bcreward = []
for file in allfiles:
with open(file, 'r') as fi:
rews = fi.read().split('\n')
rews = filter(lambda x: x != '', rews)
rews = list(map(lambda x: float(x), rews))
bcreward.extend(rews)
# Get mean and std
#print(bcreward)
mean = np.mean(bcreward)
std = np.std(bcreward)
idxcolor = 10
plt.plot([0, args.max_steps], [mean, mean],
label='BC',
color=COLORS[idxcolor])
plt.fill_between([0, args.max_steps], [mean - std, mean - std],
[mean + std, mean + std],
alpha=0.2,
color=COLORS[idxcolor])
## Get random policy
if allrandomfiles != []:
rndreward = []
for file in allrandomfiles:
with open(file, 'r') as fi:
rews = fi.read().split('\n')
rews = filter(lambda x: x != '', rews)
rews = list(map(lambda x: float(x), rews))
rndreward.extend(rews)
# Get mean and std
#print(bcreward)
mean = np.mean(rndreward)
plt.plot([0, args.max_steps], [mean, mean],
label='random',
color='gray',
linestyle='dashed')
plt.xlabel('# environment interactions', fontsize=20)
envnamehere = 'ant'
if env.lower().startswith(envnamehere):
plt.ylim(ymin=-5000, ymax=5000)
if env.lower().startswith(''):
plt.ylabel('Reward', fontsize=30)
plt.yscale(args.yscale)
plt.title(env.replace('BC','').replace('GAIL', '').replace('no', '').replace('alph', ''), \
fontsize=50)
if env.lower().startswith(envnamehere):
if args.legend != []:
if allfiles != []:
args.legend.append('BC')
plt.legend(args.legend, fontsize=30, loc='bottom right')
else:
plt.legend().set_visible(False)
#plt.ticklabel_format(useOffset=1)
plt.savefig(
'{}.png'.format(env),
bbox_inches='tight',
)
print("saved ", env)
from baselines.common import plot_util as pu
LOG_DIRS = 'logs/coinrun_500_level/'
results = pu.load_results(LOG_DIRS)
smooth_step = 50.0
fig = pu.plot_results(results,
average_group=True,
split_fn=lambda _: '',
shaded_std=False,
smooth_step=smooth_step)
pu.plt.savefig('coinrun_500_level')
from baselines.common import plot_util as pu
import matplotlib.pyplot as plt
import numpy as np
results = pu.load_results('~/logs/NewHopperCmp/')
print(len(results))
pu.plot_results(results, average_group=True, split_fn=lambda _: '')
#print(np.cumsum(results[0].monitor.l))
#plt.plot(np.cumsum(results[0].monitor.l), pu.smooth(results[0].monitor.r, radius=10))
#plt.show()
def run_cartpole_dqn(num_batches=1000,
batch_size=32,
log_dir="./logs/dqn",
seed=0):
os.makedirs(log_dir, exist_ok=True)
env = CartPoleEnv()
env.seed(seed)
torch.manual_seed(seed)
agent = CartPoleAgent(env.observation_space, env.action_space)
from baselines.bench import Monitor as BenchMonitor
env = BenchMonitor(env, log_dir, allow_early_resets=True)
train(agent, env, num_batches=num_batches, batch_size=batch_size)
return agent, env
if __name__ == "__main__":
agent, env = run_cartpole_dqn()
from baselines.common import plot_util as pu
from matplotlib import pyplot as plt
results = pu.load_results("logs")
f, ax = pu.plot_results(results)
f.savefig("logs/dqn_cartpole.png")
env = Monitor(env,
"./vid",
video_callable=lambda episode_id: True,
force=True)
visualize_it(env, agent)
