def display_sizes_iteration(prefix, **plot_kwargs):
"""
Display memory comparison
"""
pks = []
if isinstance(prefix, (str, unicode)):
prefix = [prefix]
for pre in prefix:
pks += sorted(filter(lambda x: x.startswith(pre),
listdir('../performance')),
key=lambda x: int(x.split('-')[2]),
reverse=False)
rewards_dict = OrderedDict()
x_axis = list()
for pk in sorted(pks, key=lambda x: int(x.split('-')[1])):
rewards = pickle.load(open('../performance/{}'.format(pk), 'r+'))[150:]
frame = pd.DataFrame(rewards).head(100)
key = pk.split('.')[2]
x_axis.append(int(pk.split('-')[1]))
rewards_dict.setdefault(algorithm_mapper[key], list())
rewards_dict[algorithm_mapper[key]].append(frame.sum(axis=1).mean())
results = OrderedDict()
for key in rewards_dict.keys():
results[key] = sorted(rewards_dict[key], reverse=True)
# results = rewards_dict
plot_kwargs['x_axis'] = []
for _ in x_axis:
if _ not in plot_kwargs['x_axis']:
plot_kwargs['x_axis'].append(_)
display_multiple_(results, **plot_kwargs)
def size_comparison(algorithm,
circles=200,
dump=True,
display=True,
algorithm_type='original',
prefix='',
**plot_kwargs):
"""
Algorithm comparison
Args:
algorithm(function): algorithm
circles(int): circles
dump(boolean): whether to dump result to file
display(boolean): whether to display
algorithm_type(string): algorithm type
prefix(string): prefix
"""
configs = filter(lambda x: x.startswith('sizes_comparison'),
listdir('../etc'))
rewards_dict = dict()
for config in configs:
key = 'Size-{}'.format(config.split('_')[-1]).split('.')[0]
rewards_dict[key] = simulate_with_(algorithm,
config=config,
circles=circles,
dump=dump,
algorithm_type=algorithm_type,
prefix=prefix,
fixed_theta=True)
if display:
display_multiple_(rewards_dict, **plot_kwargs)
return rewards_dict
def display_bs_comparison_by_(prefix, **plot_kwargs):
"""
Display memory comparison.
Args:
prefix(prefix): prefix string
"""
pks = sorted(filter(lambda x: x.startswith(prefix), listdir('../performance')),
key=lambda x: int(x.split('-')[2]), reverse=True)
rewards_dict = OrderedDict()
for pk in pks:
rewards = pickle.load(open('../performance/{}'.format(pk), 'r+'))
key = 'Memory-{}'.format(pk.split('-')[2])
rewards_dict[key] = rewards
display_multiple_(rewards_dict, **plot_kwargs)
def display_bs_iteration(prefix, **plot_kwargs):
"""
Display memory comparison
"""
pks = []
if isinstance(prefix, (str, unicode)):
prefix = [prefix]
for pre in prefix:
pks += sorted(filter(lambda x: x.startswith(pre), listdir('../performance')),
key=lambda x: int(x.split('-')[2]), reverse=False)
rewards_dict = OrderedDict()
x_axis = list()
for pk in sorted(pks, key=lambda x: int(x.split('-')[0].split('.')[-1])):
rewards = pickle.load(open('../performance/{}'.format(pk), 'r+'))[40:]
frame = pd.DataFrame(rewards).head(100)
key = pk.split('.')[2]
if key == 'branch_and_bound':
key = key + '_' + pk.split('.')[3]
x_axis.append(int(pk.split('-')[0].split('.')[-1]))
rewards_dict.setdefault(key, list())
rewards_dict[key].append(frame.mean(axis=1).mean())
if 'branch_and_bound_dynamic' in rewards_dict:
branch_and_bound_dynamic = np.array(rewards_dict.pop('branch_and_bound_dynamic'))
else:
branch_and_bound_dynamic = None
if 'branch_and_bound_fixed' in rewards_dict:
branch_and_bound_fixed = np.array(rewards_dict.pop('branch_and_bound_fixed'))
else:
branch_and_bound_fixed = None
delta = 0.986
if branch_and_bound_dynamic is not None and branch_and_bound_fixed is not None:
rewards_dict['branch_and_bound'] = delta * (branch_and_bound_dynamic + branch_and_bound_fixed)/2
elif branch_and_bound_dynamic is not None:
rewards_dict['branch_and_bound'] = delta * branch_and_bound_dynamic
elif branch_and_bound_fixed is not None:
rewards_dict['branch_and_bound'] = delta * branch_and_bound_fixed
else:
rewards_dict['branch_and_bound'] = delta * np.array([0] * len(rewards_dict.values()[0]))
results = OrderedDict()
for key in ['branch_and_bound', 'primal_dual_recover', 'lfu', 'lru']:
results[algorithm_mapper[key]] = sorted(rewards_dict[key])
plot_kwargs['x_axis'] = []
for _ in x_axis:
if _ not in plot_kwargs['x_axis']:
plot_kwargs['x_axis'].append(_)
display_multiple_(results, **plot_kwargs)