print num
return ax, fig
pca = True
subsample = 1
pca = True
cnn = True
locate = 95
subject_id = 7
if subject_id is 'aud':
treats = [
None
] #, 'left/auditory', 'right/auditory', 'left/visual', 'right/visual']
else:
treats = [None] #, 'face/famous','scrambled','face/unfamiliar']
nummax = len(treats)
ax = None
fig = None
num = 0
for treat in treats:
if subject_id is 'aud':
meas_dims, m, p, n_steps, total_batch_size, Wt = nn_prepro.aud_dataset(
justdims=True, cnn=cnn, locate=locate, treat=treat)
else:
meas_dims, m, p, n_steps, total_batch_size, Wt = nn_prepro.faces_dataset(
subject_id, cnn=cnn, justdims=True, locate=locate, treat=treat)
orient=None,
noise_flag=True,
selection='all',
pca=True,
subsample=1,
justdims=True,
cnn=cnn,
locate=locate,
treat=treat,
rnn=rnn,
Wt=Wt)
#print p, "Dipoles returned"
else:
meas_dims, m, p, n_steps, total_batch_size, Wt = nn_prepro.faces_dataset(
subject_id,
cnn=cnn,
justdims=True,
locate=locate,
treat=None)
meas_dims, m, p, n_steps, total_batch_size, Wt = nn_prepro.faces_dataset(
subject_id,
cnn=cnn,
justdims=True,
locate=locate,
treat=treat,
Wt=Wt)
test, val, batch_list, batches = nn_prepro.ttv(
total_batch_size,
test_frac,
val_frac,
batch_frac,
'n_layer', 'n_lstm', 'n_steps', 'train step', 'xentropy',
'rmse', 'accuracy', 'xentropy_last', 'rmse_last',
'accuracy_last'
]
subsample = 1
fname = './data/check_nn_real_relu_%s_%s_pca_%s_rand_%s.csv' % (
train_id, test_id, pca, rand_test)
with open(fname, 'a') as csvfile:
writer = csv.DictWriter(csvfile, fieldnames=fieldnames)
writer.writeheader()
for [
n_conv1, n_conv2, n_lstm, n_layer, test_frac, val_frac,
batch_frac
] in params_list:
meas_dims, m, p, n_steps, total_batch_size = nn_prepro.faces_dataset(
train_id, locate=True)
meas_dims, m, p, n_steps, total_batch_size_test = nn_prepro.faces_dataset(
test_id, locate=True)
a = np.arange(0, total_batch_size)
a_test = np.arange(0, total_batch_size_test)
#halt criteria
delta_v_err_halt = 1.
delta_err_halt = 2e-5
val_step = 50
cost = 'rmse'
cost_step = 'last'
learning_rate = 0.001
dropout = 1.0