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 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 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 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 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 main(logs='logs-arm', n_arm=3, smooth=True):
logs = glob.glob(join(logs, str(n_arm),'*'))
for log in logs:
results = pu.load_results(log)
r = results[0]
eps = np.cumsum(r[1].epoch).values
dist = r[1]['test/success_rate'].values
if smooth:
dist = pu.smooth(dist, radius=2)
plt.plot(eps, dist, label=basename(log))
plt.legend()
plt.title('Number of arms: {}'.format(n_arm))
plt.show()
def main():
parser = arg_parser()
parser.add_argument('--dir', type=str, default='logs')
parser.add_argument('--thesis', type=str, default='Online_V0')
args = parser.parse_args()
location=args.dir
thesis_dir=args.thesis
env_name_list=["Ant-v2", "HalfCheetah-v2", "Reacher-v2", "Swimmer-v2"]
#env_name_list=["HalfCheetah-v2"]
legend_name_list=["copos1", "copos1+sil_n2_l0.001", "copos1+sil_n10_l0.1",
"copos2", "copos2+sil_n2_l0.001", "copos2+sil_n10_l0.1",
"ppo", "ppo+sil_n2_l0.001", "ppo+sil_n10_l0.1",
"trpo", "trpo+sil_n2_l0.001", "trpo+sil_n10_l0.1"]
#legend_name_list=["copos1", "copos1+sil_n2_l0.001", "copos1+sil_n10_l0.1"]
for env_name in env_name_list:
dirname = '~/Desktop/logs/'+location+'/EXP_ON_V0/'+env_name
results = pu.load_results(dirname)
mn_ls, sd_ls,last_ls_ls=[],[],[]
final_txt_name="/Users/zsbjltwjj/Desktop/thesis/img/"+thesis_dir+"/"+env_name+"-final-output.txt"
for legend in legend_name_list:
result=filt(results,legend+"-")
mn, sd, last_ls = pu.table_results(result,xy_fn=pu.progress_default_xy_fn,average_group=True,split_fn=lambda _: '',
name=result[0].dirname,tp='online',freq=10)
txt_name="/Users/zsbjltwjj/Desktop/logs/"+location+"/EXP_ON_V0/"+env_name+"/"+legend+"-output.txt"
with open(txt_name, "w") as text_file:
text_file.write(str(mn)+'\n')
text_file.write(str(sd)+'\n')
for i in last_ls:
text_file.write(str(i)+' ')
# s=open(txt_name, "r")
# tmp=s.readlines()
# s.close()
mn_ls.append(mn)
sd_ls.append(sd)
last_ls_ls.append(last_ls)
#print(mn_ls)
max_idx=np.argmax(mn_ls)
with open(final_txt_name, "w") as txt_file:
bolds=[]
new_legends,new_mn_ls,new_sd_ls,new_last_ls_ls=reconstruct(max_idx,legend_name_list,mn_ls,sd_ls,last_ls_ls)
for i in range(len(new_legends)):
bold=t_test(last_ls_ls[max_idx],new_last_ls_ls[i])
bolds.append(bold)
txt_file.write(new_legends[i]+": "+str(new_mn_ls[i])+' '+str(new_sd_ls[i])+' '+str(bold)+'\n')
if any(bolds): max_bold=True
else: max_bold=False
txt_file.write("max alg: "+legend_name_list[max_idx]+": "+str(mn_ls[max_idx])+' '+str(sd_ls[max_idx])+' '+str(max_bold)+'\n')
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)
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_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_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_training(experiment_no, experiment_dir):
results = plot_util.load_results(experiment_dir, verbose=True)
results.sort(key=lambda x: x.dirname)
fig, ax = plt.subplots()
color_set = ['b', 'g', 'r', 'm', 'y', 'c']
ax.grid(color='#dddddd', linestyle='-', linewidth=1)
for i, r in enumerate(results):
i_ = i % len(color_set)
plt.plot(np.cumsum(r.monitor.l), plot_util.smooth(r.monitor.r, radius=200), color=color_set[i_])
for i, r in enumerate(results):
i_ = i % len(color_set)
plt.plot(np.cumsum(r.monitor.l), r.monitor.r, alpha=0.1, color=color_set[i_])
ax.ticklabel_format(axis='x', style='sci', scilimits=(0, 0))
ax.set_xlim(left=0)
plt.tight_layout(pad=2)
plt.xlabel('time steps')
plt.ylabel('episode_reward')
plt.title('experiment_{}'.format(experiment_no), fontsize=10)
plt.legend([os.path.basename(r.dirname) for r in results])
plt.savefig(os.path.join(experiment_dir, 'training_{}.png'.format(experiment_no)))
plt.show()
def load_data(item, path_a, window):
results = pu.load_results(path_a)
try: # for A2C
xys = [(np.array(res.progress['total_timesteps'][1:]), np.array(res.progress[item][1:])) for res in results]
except: # for TNPG
xys = [(np.array(res.progress['TimestepsSoFar']), np.array(res.progress[item])) for res in results]
origxs = [xy[0] for xy in xys]
print([x[-1] for x in origxs])
low = max(x[0] for x in origxs)
high = min(x[-1] for x in origxs)
resample = len(origxs[0])
ts = np.linspace(low, high, resample)+1
ys = []
for (x, y) in xys:
ys.append(symmetric_ema(x, y, low, high, resample, decay_steps=window)[1])
val = np.array(ys).T
mean_val = np.mean(val, axis=1)
sd_val = np.std(val, axis=1)
med_val = np.median(val, axis=1)
q1_val = np.percentile(val, 25, axis=1)
q3_val = np.percentile(val, 75, axis=1)
return {'ts':ts, 'median':med_val, 'q1':q1_val, 'q3':q3_val, 'mean':mean_val, 'std':sd_val}
from baselines.common import plot_util as pu
mujoco_results = pu.load_results(
'/home/developer/logs/her_pyrobot_push_mujoco/250k_v2')
#results = pu.load_results('/home/developer/logs/her_pyrobot_reach/joint_100k_v4')
import matplotlib.pyplot as plt
import numpy as np
mujoco_r = mujoco_results[0]
mujoco_arr = np.array(mujoco_r.progress)
epoch = mujoco_arr[:, 0]
mujoco_test_success_rate = mujoco_arr[:, 7]
mujoco_train_success_rate = mujoco_arr[:, 9]
plt.suptitle(
'LocoBot Push Task Trained in MuJoCo with 250k timesteps(100 epoches)~')
plt.title('Training in Mujoco')
plt.plot(epoch,
pu.smooth(mujoco_train_success_rate, radius=10),
label="mujoco_train")
plt.plot(epoch,
pu.smooth(mujoco_test_success_rate, radius=10),
label="mujoco_test")
plt.xlabel('epoches')
plt.ylabel('success_rate')
plt.legend()
#plt.plot(np.cumsum(r.monitor.l), pu.smooth(r.monitor.r, radius=10))
plt.show()
plot_envs = ["Qbert", "SpaceInvaders", 'Breakout', "BeamRider"]
pre = ""
means_cache = {}
if os.path.exists(pre + "mean_reward_cache.pkl"):
means_cache = pickle.load(open(pre + "mean_reward_cache.pkl", 'rb'))
for i, hp in tqdm(hps):
log_dir = hp['log_dir']
if not os.path.exists(log_dir):
continue
if log_dir in means_cache:
r = means_cache[log_dir]
else:
try:
result = pu.load_results(log_dir)
if pre == "":
r = result[0].monitor['r'].mean()
else:
r = result[0].monitor['r'].tail(1000).mean()
except Exception as e:
if log_dir in dont_cache:
print(e, log_dir)
continue
if log_dir not in dont_cache:
means_cache[log_dir] = r
if not np.isfinite(r):
#print("reward for",i, log_dir,"?", r)
continue
rewards.append(r)
from baselines.common import plot_util as pu
results = pu.load_results('~/Documents/openai-results', verbose=True)
# results = pu.load_results('/tmp')
import matplotlib.pyplot as plt
import numpy as np
lines = ["-", "--", ":", "-."]
markers = ['o', 'x', '+', '^']
colors = ['#000000', '#222222', '#444444', '#666666']
def average_graph():
min_number_episodes = 10000
for i in range(len(results)):
result = results[i]
rewards = result.monitor.r
number_episodes = len(rewards)
print('Results {}: number of episodes = {}'.format(i, number_episodes))
if (number_episodes < min_number_episodes):
min_number_episodes = number_episodes
total_reward_per_episode = [0] * min_number_episodes
for i in range(len(results)):
result = results[i]
rewards = result.monitor.r
for j in range(min_number_episodes):
f, ax = pu.plot_results(
results,
xy_fn=xy_fn,
split_fn=lambda _: "",
average_group=True,
shaded_err=False,
)
f.savefig(file)
if __name__ == "__main__":
# [run_cartpole_dqn(log_dir="logs/dqn-%d" % k, seed=k) for k in range(3)]
#
# [
# run_cartpole_reinforce(
# RLparams(num_games=200, seed=k), log_dir="logs/reinforce-%d" % k
# )
# for k in range(3)
# ]
[
run_cartpole_a2c(
A2CParams(lr=0.01, num_rollout_steps=32, num_batches=1000, seed=k),
log_dir="logs/a2c-%d" % k,
) for k in range(3)
]
results = pu.load_results("logs")
plot_save_results(lr_fn, "logs/steps_rewards.png")
plot_save_results(tr_fn, "logs/time_rewards.png")
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 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 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
# Parameters
dir_name = 'openai-2018-11-06-21-19-32-539241'
game_name = 'Pong'
results = pu.load_results(dir_name)
r = results[0]
plt.plot(np.cumsum(r.monitor.l), pu.smooth(r.monitor.r, radius=10))
plt.title(game_name)
plt.xlabel('# of timesteps')
plt.ylabel('Reward')
#plt.show()
plt.savefig(game_name + '_rewards.png', bbox_inches='tight')
# Copyright (C) 2019 Istituto Italiano di Tecnologia (IIT)
# This software may be modified and distributed under the terms of the
# LGPL-2.1+ license. See the accompanying LICENSE file for details.
import os, inspect
currentdir = os.path.dirname(
os.path.abspath(inspect.getfile(inspect.currentframe())))
print(currentdir)
parentdir = os.path.dirname(os.path.dirname(currentdir))
os.sys.path.insert(0, parentdir)
import baselines
from baselines.common import plot_util as pu
import matplotlib.pyplot as plt
import numpy as np
results = pu.load_results('')
r = results[0]
print(r.monitor.r)
plt.plot(np.cumsum(r.monitor.l), pu.smooth(r.monitor.r, radius=10))
plt.show()
import argparse
from baselines.common import plot_util as pu
parser = argparse.ArgumentParser()
parser.add_argument('path', help='an integer for the accumulator')
args = parser.parse_args()
exp_prefix = '/home/murtaza/research/baselines/logs/'
results = pu.load_results(exp_prefix + args.path)
import matplotlib.pyplot as plt
import numpy as np
r = results[0]
plt.plot(r.progress.epoch, r.progress['test/success_rate'])
plt.show()
from baselines.common import plot_util as pu
import matplotlib.pyplot as plt
import numpy as np
import json
import os
if not os.path.exists('pics/'):
os.makedirs('pics/')
results = pu.load_results('logs/', verbose=True)
for r in results:
with open("{}/0.0.monitor.csv".format(r.dirname), "r") as f:
info = f.readline()
meta = json.loads(info[1:])
game = meta['env_id']
fig = plt.figure()
plt.plot(np.cumsum(r.monitor.l), pu.smooth(r.monitor.r, radius=10))
fig.suptitle(game, fontsize=18)
plt.savefig('pics/{}.png'.format(game))
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 save_plot(path, seed):
results = pu.load_results(dirs + '%s/%s/' % (path, seed))
r = results[0]
plt.plot(r.progress.steps, r.progress["mean 100 episode reward"])
plt.savefig(dirs + '%s/%s/plot.png' % (path, seed))
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
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()
