p_test = 0
p_rm_ob_enc_test = p_test
p_rm_ob_rcl_test = p_test
similarity_max_test = .9
similarity_min_test = 0
fix_cond = None
n_examples_test = 256
subj_id = 0
p = P(
exp_name=exp_name,
sup_epoch=supervised_epoch,
n_param=n_param,
n_branch=n_branch,
pad_len=pad_len_load,
def_prob=def_prob,
n_def_tps=n_def_tps,
enc_size=enc_size,
attach_cond=attach_cond,
penalty=penalty_train,
p_rm_ob_enc=p_rm_ob_enc_load,
p_rm_ob_rcl=p_rm_ob_rcl_load,
)
# create logging dirs
log_path, log_subpath = build_log_path(subj_id,
p,
log_root=log_root,
mkdir=False,
verbose=True)
# init env
env_data = load_env_metadata(log_subpath)
for subj_id in subj_ids:
print(f'\nsubj_id = {subj_id}: ', end='')
for fix_cond in all_conds:
print(f'{fix_cond} ', end='')
np.random.seed(subj_id)
p = P(
exp_name=exp_name,
sup_epoch=supervised_epoch,
n_param=n_param,
n_branch=n_branch,
pad_len=pad_len_load,
enc_size=enc_size,
n_event_remember=n_event_remember,
penalty=penalty_train,
penalty_random=penalty_random,
penalty_onehot=penalty_onehot,
penalty_discrete=penalty_discrete,
normalize_return=normalize_return,
p_rm_ob_enc=p_rm_ob_enc_load,
p_rm_ob_rcl=p_rm_ob_rcl_load,
n_hidden=n_hidden,
n_hidden_dec=n_hidden_dec,
lr=learning_rate,
eta=eta,
)
# init env
task = SequenceLearning(
n_param=p.env.n_param,
n_branch=p.env.n_branch,
pad_len=pad_len_test,
p_rm_ob_enc=p_rm_ob_enc_test,
np.random.seed(seed_val)
torch.manual_seed(seed_val)
p = P(exp_name=exp_name,
subj_id=subj_id,
sup_epoch=supervised_epoch,
n_param=n_param,
n_branch=n_branch,
pad_len=pad_len,
def_prob=def_prob,
n_def_tps=n_def_tps,
enc_size=enc_size,
dict_len=dict_len,
n_event_remember=n_event_remember,
penalty=penalty,
penalty_random=penalty_random,
penalty_discrete=penalty_discrete,
penalty_onehot=penalty_onehot,
normalize_return=normalize_return,
attach_cond=attach_cond,
p_rm_ob_enc=p_rm_ob_enc,
p_rm_ob_rcl=p_rm_ob_rcl,
n_hidden=n_hidden,
n_hidden_dec=n_hidden_dec,
lr=learning_rate,
eta=eta,
cmpt=cmpt,
repeat_query=repeat_query)
# init env
task = SequenceLearning(
n_param=p.env.n_param,
n_branch = 4
n_param = 16
enc_size = 16
n_event_remember = 2
similarity_max_test = .9
similarity_min_test = 0
n_examples_test = 256
p = P(
exp_name=exp_name,
sup_epoch=supervised_epoch,
n_param=n_param,
n_branch=n_branch,
pad_len=pad_len_load,
enc_size=enc_size,
n_event_remember=n_event_remember,
def_prob=def_prob,
n_def_tps=n_def_tps,
penalty=penalty_train,
penalty_random=penalty_random,
p_rm_ob_enc=p_rm_ob_enc_load,
p_rm_ob_rcl=p_rm_ob_rcl_load,
cmpt=comp_val,
)
'''load data'''
lca_param = {ptest: None for ptest in penaltys_test}
auc = {ptest: None for ptest in penaltys_test}
acc = {ptest: None for ptest in penaltys_test}
mis = {ptest: None for ptest in penaltys_test}
dk = {ptest: None for ptest in penaltys_test}
ma_lca = defaultdict()
ma_cosine = defaultdict()
penalty_random = 1
supervised_epoch = 600
# loading params
pad_len_load = -1
p_rm_ob_enc_load = .3
p_rm_ob_rcl_load = 0
# testing params
pad_len_test = 0
n_examples_test = 256
slience_recall_time = None
test_params = [fix_penalty, pad_len_test, slience_recall_time]
# store params
p = P(
exp_name=exp_name, sup_epoch=supervised_epoch,
n_param=n_param, n_branch=n_branch, pad_len=pad_len_load,
enc_size=enc_size, penalty=penalty_train, penalty_random=penalty_random,
cmpt=cmpt, p_rm_ob_enc=p_rm_ob_enc_load, p_rm_ob_rcl=p_rm_ob_rcl_load,
)
# init a dummy task
task = SequenceLearning(n_param=p.env.n_param, n_branch=p.env.n_branch)
dacc = np.zeros((n_subjs, n_param, T))
Yob_all, Yhat_all = [], []
o_keys_p1_all, o_keys_p2_all = [], []
for i_s in range(n_subjs):
# create logging dirs
np.random.seed(i_s)
log_path, log_subpath = build_log_path(i_s, p, log_root, mkdir=False)
test_data_fname = get_test_data_fname(n_examples_test, fix_cond=fix_cond)
test_data_dir, _ = get_test_data_dir(log_subpath, epoch_load, test_params)
fpath = os.path.join(test_data_dir, test_data_fname)