epochs=10,
train_batches_per_epoch=10,
# ---- sparsity related
target_final_density=0.2,
sparse_start=None,
sparse_end=None,
on_perc=1.0,
# ---- optimizer related
optim_alg="SGD",
learning_rate=0.1,
weight_decay=0,
)
# run
tune_config = dict(
name=os.path.basename(__file__).replace(".py", "") + "_lt",
num_samples=1,
local_dir=os.path.expanduser("~/nta/results"),
checkpoint_freq=0,
checkpoint_at_end=False,
resources_per_trial={
"cpu": 1,
"gpu": 1
},
verbose=0,
)
run_ray(tune_config, exp_config, fix_seed=True)
# 10/31 - ran script, working ok, results as expected
hebbian_prune_perc=tune.grid_search([0, 0.1, 0.2, 0.3, 0.4, 0.5]), # 6
weight_prune_perc=tune.grid_search([0, 0.1, 0.2, 0.3, 0.4, 0.5]), # 6
pruning_es=False,
pruning_active=True,
hebbian_grow=tune.grid_search([True, False]), # 2
# additional validation
test_noise=True,
noise_level=0.15, # test with more agressive noise
# debugging
debug_weights=True,
debug_sparse=True,
)
# ray configurations
tune_config = dict(
name=__file__.replace(".py", "") + "_eval2",
num_samples=1,
local_dir=os.path.expanduser("~/nta/results"),
checkpoint_freq=0,
checkpoint_at_end=False,
stop={"training_iteration": 30},
resources_per_trial={
"cpu": 1,
"gpu": 1
},
loggers=DEFAULT_LOGGERS,
verbose=1,
)
run_ray(tune_config, base_exp_config)
batch_size_test=1024,
# 1- random search baseline
learning_rate=tune.sample_from(
lambda spec: np.random.uniform(0.0001, 0.2)),
on_perc=tune.sample_from(lambda spec: np.random.uniform(0.1, 1.0)),
momentum=tune.sample_from(lambda spec: np.random.uniform(0, 1.0)),
weight_decay=tune.sample_from(lambda spec: np.random.uniform(0, 0.1)),
# 2- sigopt extra parameters
# experiment_type="SigOpt",
# params_space=sigopt_params_space,
# performance_metric="val_acc"
)
# run
tune_config = dict(
# name=__file__,
name="sigopt_test3",
num_samples=300,
local_dir=os.path.expanduser("~/nta/results"),
checkpoint_freq=0,
checkpoint_at_end=False,
stop={"training_iteration": 20},
resources_per_trial={
"cpu": 0,
"gpu": 0.2
},
verbose=2,
)
run_ray(tune_config, exp_config)
