def load_results(exp_cfgs):
exp_cfgs = exp_cfgs[0] + exp_cfgs[1]
sources, targets = {}, {}
for i, exp_cfg in enumerate(exp_cfgs):
cwd = os.getcwd()
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
results_hub = hub.ResultsHub(batch, kind='cov')
src_data = results_hub.data()
assert len(src_data) == 1
results = {'exp_cfg': exp_cfg}
results.update(src_data[0])
if len(exp_cfg.exp.key[1]) == 0:
sources[(exp_cfg.exp.env_name,
exp_cfg.exp.explorer_name)] = results
else:
if len(src_data[0]['avg']) == 0:
print('MISSING: {}'.format(exp_cfg.exp.key))
else:
targets[exp_cfg.exp.key] = results
os.chdir(cwd)
return sources, targets
def load_data(kind):
cwd = os.getcwd()
Ns = [1000, 2000, 5000, 10000]
ps, avgs, stds = ({(env, d, N): [] for env in env_names for d in disturbs for N in Ns},
{(env, d, N): {} for env in env_names for d in disturbs for N in Ns},
{(env, d, N): {} for env in env_names for d in disturbs for N in Ns})
ds = [0.001, 0.05, 0.5]
env_keys = []
min_avgs, min_stds = {}, {}
for i, exp_cfg in enumerate(exp_cfgs):
if exp_cfg.exploration.explorer.ex_1.learner.m_disturb in ds:
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
results_hub = hub.ResultsHub(batch, kind=kind)
data = results_hub.data()[0]
d = exp_cfg.exploration.explorer.ex_1.learner.m_disturb
p = exp_cfg.exploration.explorer.weights[1][0]
for N in Ns:
env = exp_cfg.exp.env_name
index = data['ticks'].index(N)
env_key = (env, d, N)
ps[env_key].append(p)
avgs[env_key][p] = data['avg'][index]
stds[env_key][p] = data['std'][index]
if p == 0:
env_keys.append(env_key)
for t in data['ticks']:
env_name = exp_cfg.exp.env_name
env_key = (env_name, d, t)
min_avgs.setdefault(env_key, float('inf'))
index = data['ticks'].index(t)
if min_avgs[env_key] >= data['avg'][index]:
min_avgs[env_key] = data['avg'][index]
min_stds[env_key] = data['std'][index]
for env_key in env_keys:
ps[env_key] = sorted(ps[env_key])
avgs[env_key] = np.array([avgs[env_key][p] for p in ps[env_key]])
stds[env_key] = np.array([stds[env_key][p] for p in ps[env_key]])
os.chdir(cwd)
return env_keys, ps, avgs, stds, min_avgs, min_stds
def load_data(kind):
cwd = os.getcwd()
Ns = [1000, 2000, 5000]
ps, avgs, stds = ({(env, d, N): [] for env in env_names for d in disturbs for N in Ns},
{(env, d, N): {} for env in env_names for d in disturbs for N in Ns},
{(env, d, N): {} for env in env_names for d in disturbs for N in Ns})
ds = [0.001, 0.05, 0.5]
env_keys = []
min_avgs, min_stds = {}, {}
for i, exp_cfg in enumerate(exp_cfgs):
if exp_cfg.exploration.explorer.ex_1.learner.m_disturb in ds:
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
results_hub = hub.ResultsHub(batch, kind=kind)
data = results_hub.data()[0]
d = exp_cfg.exploration.explorer.ex_1.learner.m_disturb
p = exp_cfg.exploration.explorer.weights[1][0]
for N in Ns:
env = exp_cfg.exp.env_name
index = data['ticks'].index(N)
env_key = (env, d, N)
ps[env_key].append(p)
avgs[env_key][p] = data['avg'][index]
stds[env_key][p] = data['std'][index]
if p == 0:
env_keys.append(env_key)
for t in data['ticks']:
env_name = exp_cfg.exp.env_name
env_key = (env_name, d, t)
min_avgs.setdefault(env_key, float('inf'))
index = data['ticks'].index(t)
if min_avgs[env_key] >= data['avg'][index]:
min_avgs[env_key] = data['avg'][index]
min_stds[env_key] = data['std'][index]
for env_key in env_keys:
ps[env_key] = sorted(ps[env_key])
avgs[env_key] = np.array([avgs[env_key][p] for p in ps[env_key]])
stds[env_key] = np.array([stds[env_key][p] for p in ps[env_key]])
os.chdir(cwd)
return env_keys, ps, avgs, stds, min_avgs, min_stds
def coverage_graphs(expcfgs_levels, dest='tmp.html', n_graphs=3):
cwd = os.getcwd()
graphs.output_file(dest)
red = '#DF6464'
for level in expcfgs_levels:
for i, exp_cfg in enumerate(level):
n = 0
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
data_hub = hub.DataHub(batch, sensory_only=True)
datas = data_hub.data()
for j, data in enumerate(datas):
if n_graphs is None or n < n_graphs:
for N in [200, 1000]:
print(exp_cfg.exploration.explorer)
n_reuse = exp_cfg.exploration.explorer.eras[0]
s_vectors = [
tools.to_vector(s_signal, data.s_channels)
for s_signal in data.s_signals
][:N]
graphs.coverage(data.s_channels,
exp_cfg.testscov.buffer_size,
s_vectors=s_vectors,
swap_xy=False,
title_text_font_size='6pt',
title='{} {}'.format(
exp_cfg.exp.key, j))
graphs.hold(True)
graphs.spread(data.s_channels,
s_vectors=s_vectors[n_reuse:],
swap_xy=False,
e_radius=2.0)
graphs.hold(True)
graphs.spread(data.s_channels,
s_vectors=s_vectors[:n_reuse],
swap_xy=False,
e_radius=2.0,
e_color=red)
n += 1
os.chdir(cwd)
def dovecot_replay(exp_cfg, rep, collision_only=True, headless=False):
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
data_hub = hub.DataHub(batch,
folder='/Users/fabien/research/data.hard/',
verbose=True)
job_data = data_hub.data()
job_data = job_data[rep]
assert job_data.job.cfg.job.rep == rep
job_cfg = job_data.jobcfg._deepcopy()
job_cfg.job._freeze(False)
env_cfg = job_cfg.job.env
env_cfg = envs.catalog[env_name]
if headless:
env_cfg.execute.simu.ppf = 200
env_cfg.execute.simu.headless = True
else:
env_cfg.execute.simu.ppf = 1
env_cfg.execute.simu.headless = False
env = environments.Environment.create(env_cfg)
raw_input()
assert job_data.observations is not None, "motor signals (observations) could not be loaded (datafiles probably missing)."
print('obs from file: {}{}{}'.format(gfx.green,
job_cfg.exploration.hardware.datafile,
gfx.end))
for step, (m_signal, s_signal) in enumerate(job_data.observations):
if not collision_only or s_signal['push_saliency'] != 0:
# if s_signal['x'] > 100.0 and s_signal['y'] > 100.0:
#if step in [896, 786, 648, 414]:
#assert s_signal['push_saliency'] > 999
print('{:04d} {}{}{} [recorded]'.format(step, gfx.cyan, s_signal,
gfx.end))
feedback = env.execute(m_signal)
print(' {}'.format(feedback['s_signal']))
raw_input()
import dotdot
import graphs
from fig3_8_unreach_graph import ratios, avgs, stds
from figB_4_adapt_reach_grid_cluster import expcfgs_levels, windows, RESS
#Ns = [100, 1000, 2000, 5000, 10000]
Ns = [2000, 10000]
W = windows[0]
cwd = os.getcwd()
adapt_avgs = {}
adapt_stds = {}
for i, exp_cfg in enumerate(expcfgs_levels[0]):
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
data = hub.ResultsHub(batch, kind='nn').data()[0]
window = exp_cfg.exploration.explorer.ex_1.window
res = exp_cfg.exploration.explorer.ex_1.res
for N in Ns:
print(data['ticks'])
index = data['ticks'].index(N)
adapt_avgs[(window, N, res)] = data['avg'][index]
adapt_stds[(window, N, res)] = data['std'][index]
os.chdir(cwd)
# for N in Ns:
# for w in windows:
def dovecot_replay(exp_cfg, rep, collision_only=True, headless=False):
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
data_hub = hub.DataHub(batch, folder='/Users/fabien/research/data.sim/', verbose=True)
job_data = data_hub.data()
job_data = job_data[rep]
assert job_data.job.cfg.job.rep == rep
job_cfg = job_data.jobcfg._deepcopy()
job_cfg.job._freeze(False)
env_cfg = job_cfg.job.env
env_cfg = envs.catalog[env_name]
env_cfg.execute.prefilter = False
env_cfg.execute.simu.calibr_check = False
if headless:
env_cfg.execute.simu.ppf = 200
env_cfg.execute.simu.headless = True
else:
env_cfg.execute.simu.ppf = 1
env_cfg.execute.simu.headless = False
env = environments.Environment.create(env_cfg)
raw_input()
assert job_data.observations is not None, "motor signals (observations) could not be loaded (datafiles probably missing)."
print('obs from file: {}{}{}'.format(gfx.green, job_cfg.exploration.hardware.datafile, gfx.end))
for step, (expl, fback) in enumerate(job_data.explorations):
s_vector = environments.tools.to_vector(fback['s_signal'], job_data.s_channels)
print('{}:{}: {}'.format(step, s_vector[-1], expl['from']))
step = -1
while True:
print('choose a step:')
inp = raw_input()
if inp == '':
step += 1
elif inp == '=':
step += 0 # no change
else:
step = int(inp)
expl, fback = job_data.explorations[step]
m_signal = expl['m_signal']
s_signal = fback['s_signal']
# for step, (expl, fback) in enumerate(job_data.explorations):
# m_signal = expl['m_signal']
# s_signal = fback['s_signal']
# print(m_signal)
# m_signal = environments.tools.random_signal(job_data.m_channels)
# print(m_signal)
print('{:04d} {}'.format(step, expl['from']))
# if step in [50, 49, 7, 19, 15]:
# if (expl['from'] == 'goal.babbling'):
# if s_signal['push_saliency'] == 0:
# if (not collision_only or s_signal['push_saliency'] == 0):
# if s_signal['x'] > 100.0 and s_signal['y'] > 100.0:
#if step in [896, 786, 648, 414]:
#assert s_signal['push_saliency'] > 999
# for _ in range(5):
print('{:04d} {}{}{} [recorded]'.format(step, gfx.cyan, s_signal, gfx.end))
feedback = env.execute(m_signal)
print(' {}'.format(feedback['s_signal']))
import graphs
from fixed_graph import load_data, disturbs
from ddmab_cluster import expcfgs_levels, cfg
Ns = [5000]
mbratio_env_keys, ps, avgs, stds, min_avgs, min_stds = load_data('cov')
cwd = os.getcwd()
adapt_avgs = {}
adapt_stds = {}
for i, exp_cfg in enumerate(expcfgs_levels[0]):
batch = jobfactory.make_jobgroup([exp_cfg])
os.chdir(os.path.expanduser(exp_cfg.meta.rootpath))
data = hub.ResultsHub(batch, kind=KIND).data()[0]
window = exp_cfg.exploration.explorer.window
d = exp_cfg.exploration.explorer.ex_1.learner.m_disturb
# for N in cfg.testsnn.ticks:
for N in Ns:
index = data['ticks'].index(N)
adapt_avgs[(d, N)] = data['avg'][index]
adapt_stds[(d, N)] = data['std'][index]
os.chdir(cwd)