ppn=12,
cpp=2,
gpu_set="0,1,2,3",
pmem=5000,
project="rpp-bengioy",
wall_time="71hours",
cleanup_time="5mins",
slack_time="5mins",
n_repeats=2,
copy_locally=True,
config=dict(max_steps=120000,
patience=0,
curriculum=[dict()]))
durations = dict(
long=copy_update(run_kwargs),
short=dict(
wall_time="180mins",
gpu_set="0",
ppn=4,
n_repeats=4,
distributions=None,
config=dict(max_steps=3000,
render_step=500,
eval_step=100,
display_step=100,
stage_steps=600,
curriculum=[dict()]),
),
test_load=dict(
wall_time="180mins",
run_kwargs = dict(
max_hosts=1,
ppn=4,
cpp=4,
gpu_set="0,1",
pmem=5000,
project="rpp-bengioy",
wall_time="96hours",
cleanup_time="5mins",
slack_time="5mins",
n_repeats=4,
)
durations = dict(
long=copy_update(run_kwargs),
restart10=copy_update(
run_kwargs,
wall_time="75hours",
cpp=4,
pmem=5000,
ppn=1,
n_repeats=1,
gpu_set="0",
config=dict(
seed=100,
restart_steps="1:120000",
experiment_restart_path=
"/scratch/e2crawfo/dps_data/parallel_experiments_run/aaai_2020_silot/shapes/run/run_env=big-shapes_max-shapes=10_alg=shapes-silot_duration=long_2019_07_30_16_55_21_seed=0/experiments",
prepare_func=silot_shapes_restart_prepare_func,
),
run_kwargs = dict(max_hosts=2,
ppn=8,
cpp=2,
gpu_set="0,1,2,3",
pmem=10000,
project="rpp-bengioy",
wall_time="71hours",
cleanup_time="5mins",
slack_time="5mins",
n_repeats=1,
copy_locally=True,
config=dict(max_steps=200000, render_step=1000000))
durations = dict(
long=copy_update(run_kwargs),
medium=copy_update(
run_kwargs,
wall_time="6hours",
config=dict(stage_steps=3000, max_steps=12000),
),
short=dict(
wall_time="180mins",
gpu_set="0",
ppn=4,
n_repeats=4,
distributions=None,
config=dict(max_steps=3000,
render_step=500,
eval_step=100,
display_step=100,
gpu_set="0",
wall_time="6hours",
n_repeats=1,
distributions=None,
config=dict(do_train=False,
n_train=32,
n_val=1008,
get_updater=DummyUpdater,
render_hook=None,
curriculum=[
dict(min_shapes=i, max_shapes=i)
for i in range(1, 36)
])),
)
durations['long_restart'] = copy_update(durations['long'], ppn=1, gpu_set="0")
config = basic_config.copy()
config.update(env_configs['big_shapes'])
config.update(alg_configs['shapes_silot'])
config.batch_size = 8
config.n_prop_objects = 36
config.update(max_shapes=args.max_shapes, small=args.is_small)
run_experiment("shapes_silot",
config,
"silot on shapes.",
name_variables="max_shapes,small",
durations=durations)
{'cc_threshold': 0.8236294388771057,
'cosine_threshold': 0.9616384506225586},
{'cc_threshold': 0.8236294388771057,
'cosine_threshold': 0.9616384506225586}
],
}
alg_configs['shapes_baseline_test'] = alg_configs['shapes_baseline'].copy(
do_train=False,
cc_threshold=None,
cosine_threshold=None,
)
alg_configs['shapes_baseline_AP'] = alg_configs['shapes_baseline_test'].copy(
curriculum=[
copy_update(v, min_shapes=i, max_shapes=i)
for i, v in zip(shapes_baseline_n_shapes, shapes_baseline_values['AP'])]
)
alg_configs['shapes_baseline_count_1norm'] = alg_configs['shapes_baseline_test'].copy(
curriculum=[
copy_update(v, min_shapes=i, max_shapes=i)
for i, v in zip(shapes_baseline_n_shapes, shapes_baseline_values['count_1norm'])]
)
alg_configs['shapes_baseline_mota'] = alg_configs['shapes_baseline_test'].copy(
curriculum=[
copy_update(v, min_shapes=i, max_shapes=i)
for i, v in zip(shapes_baseline_n_shapes, shapes_baseline_values['MOT:mota'])]
)