output_dir = "3d_conv_3l_" + str(begin_subject) + "_" + str(
end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
output_file = "3d_conv_3l_" + str(begin_subject) + "_" + str(
end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
os.system("mkdir " + result_dir + "/" + output_dir + " -p")
###########################################################################
# build network
###########################################################################
# instance cnn class
cnn_3d = cnn()
# input placeholder
X = tf.placeholder(
tf.float32,
shape=[None, input_depth, input_height, input_width, input_channel_num],
name='X')
Y = tf.placeholder(tf.float32, shape=[None, num_labels], name='Y')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# first CNN layer
conv_1 = cnn_3d.apply_conv3d(X, kernel_depth_1st, kernel_height_1st,
kernel_width_1st, input_channel_num,
conv_channel_num, kernel_stride)
# pool_1 = cnn_3d.apply_max_pooling3d(conv_1, pooling_depth, pooling_height, pooling_width, pooling_stride)
) + "_fc_" + str(n_fc_in) + "_RNN" + str(n_lstm_layers) + "_fc_" + str(
n_fc_out) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d') + "_hs_" + str(n_hidden_state))
output_file = "win_" + str(window_size) + "_" + str(begin_subjec) + "_" + str(
end_subject
) + "_fc_" + str(n_fc_in) + "_RNN" + str(n_lstm_layers) + "_fc_" + str(
n_fc_out) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d') + "_hs_" + str(n_hidden_state))
os.system("mkdir " + result_dir + "/" + output_dir + " -p")
###########################################################################
# build network
###########################################################################
# instance cnn class
rnn = cnn()
# input placeholder
X = tf.placeholder(tf.float32,
shape=[None, n_time_step, n_input_ele],
name='X')
Y = tf.placeholder(tf.float32, shape=[None, num_labels], name='Y')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
####################### input fully connected layer ##########################
# X ==> [batch_size, n_time_step, n_input_ele]
shape = X.get_shape().as_list()
# X_flat ==> [batch_size*n_time_step, n_input_ele]
X_flat = tf.reshape(X, [-1, shape[2]])
begin_subject) + "_" + str(end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
output_file = str(cnn_dimension) + "d_conv_3l_" + str(
begin_subject) + "_" + str(end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
os.system("mkdir " + result_dir + "/" + output_dir + " -p")
###########################################################################
# build network
###########################################################################
# instance cnn class
if (cnn_dimension == 1):
cnn_1d = cnn()
elif (cnn_dimension == 2):
cnn_2d = cnn()
elif (cnn_dimension == 3):
cnn_3d = cnn()
else:
print(
"******************* FATAL unknown dimension ! ! ! ********************"
)
sys.exit()
# input placeholder
if (cnn_dimension == 1):
X = tf.placeholder(tf.float32,
shape=[None, input_width, input_channel_num],
name='X')
learning_rate = 1e-5 # set maximum traing epochs training_epochs = 110 # set batch size batch_size = 10 # set dropout probability dropout_prob = 0.5 # set train batch number per epoch batch_num_per_epoch = features_train.shape[0]//batch_size # instance cnn class cnn_2d = cnn(padding='VALID') # input placeholder X = tf.placeholder(tf.float32, shape=[None, input_height, input_width, input_channel_num], name = 'X') Y = tf.placeholder(tf.float32, shape=[None, num_labels], name = 'Y') train_phase = tf.placeholder(tf.bool, name = 'train_phase') keep_prob = tf.placeholder(tf.float32, name='keep_prob') # first CNN layer conv_1 = cnn_2d.apply_conv2d(X, kernel_height_1st, kernel_width_1st, input_channel_num, conv_channel_num, kernel_stride, train_phase) pool_1 = cnn_2d.apply_max_pooling(conv_1, pooling_height_1st, pooling_width_1st, pooling_stride_1st) pool1_shape = pool_1.get_shape().as_list() pool1_flat = tf.reshape(pool_1, [-1, pool1_shape[1]*pool1_shape[2]*pool1_shape[3]]) fc_drop = tf.nn.dropout(pool1_flat, keep_prob)
# set maximum traing epochs
training_epochs = 200
# set batch size
batch_size = 10
# set dropout probability
dropout_prob = 0.5
# set train batch number per epoch
batch_num_per_epoch = train_x.shape[0] // batch_size
# instance cnn class
padding = 'VALID'
cnn_2d = cnn(padding=padding)
# input placeholder
X = tf.placeholder(tf.float32,
shape=[None, input_height, input_width, input_channel_num],
name='X')
Y = tf.placeholder(tf.float32, shape=[None, num_labels], name='Y')
train_phase = tf.placeholder(tf.bool, name='train_phase')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# first CNN layer
conv_1 = cnn_2d.apply_conv2d(X, kernel_height_1st, kernel_width_1st,
input_channel_num, conv_channel_num,
kernel_stride, train_phase)
print("conv 1 shape: ", conv_1.get_shape().as_list())
pool_1 = cnn_2d.apply_max_pooling(conv_1, pooling_height_1st,
regularization_method = 'dropout+l2'
else:
regularization_method = 'dropout'
# result output
result_dir = "/home/dadafly/experiment_result/eeg_physiobank_cnn_result"
output_dir = "1d_conv_3l_"+str(begin_subjec)+"_"+str(end_subject)+"_fc_"+str(fc_size)+"_"+regularization_method+"_"+str(format(train_test_split*100, '03d'))
output_file = "1d_conv_3l_"+str(begin_subjec)+"_"+str(end_subject)+"_fc_"+str(fc_size)+"_"+regularization_method+"_"+str(format(train_test_split*100, '03d'))
os.system("mkdir "+result_dir+"/"+output_dir+" -p")
###########################################################################
# build network
###########################################################################
# instance cnn class
cnn_1d = cnn()
# input placeholder
X = tf.placeholder(tf.float32, shape=[None, input_width, input_channel_num], name = 'X')
Y = tf.placeholder(tf.float32, shape=[None, num_labels], name = 'Y')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# first CNN layer
conv_1 = cnn_1d.apply_conv1d(X, kernel_width_1st, input_channel_num, conv_channel_num, kernel_stride)
# pool_1 = cnn_1d.apply_max_pooling(conv_1, pooling_height, pooling_width_1st, pooling_stride)
# second CNN layer
conv_2 = cnn_1d.apply_conv1d(conv_1, kernel_width_2nd, conv_channel_num, conv_channel_num*2, kernel_stride)
# pool_2 = apply_max_pooling(conv_2, pooling_height, pooling_width_2nd, pooling_stride)
# third CNN layer
output_dir = "2d_conv_3l_" + str(begin_subject) + "_" + str(
end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
output_file = "2d_conv_3l_" + str(begin_subject) + "_" + str(
end_subject) + "_fc_" + str(
fc_size) + "_" + regularization_method + "_" + str(
format(train_test_split * 100, '03d'))
os.system("mkdir " + result_dir + "/" + output_dir + " -p")
###########################################################################
# build network
###########################################################################
# instance cnn class
cnn_2d = cnn()
# input placeholder
X = tf.placeholder(tf.float32,
shape=[None, input_height, input_width, input_channel_num],
name='X')
Y = tf.placeholder(tf.float32, shape=[None, num_labels], name='Y')
keep_prob = tf.placeholder(tf.float32, name='keep_prob')
# first CNN layer
conv_1 = cnn_2d.apply_conv2d(X, kernel_height_1st, kernel_width_1st,
input_channel_num, conv_channel_num,
kernel_stride)
# pool_1 = cnn_1d.apply_max_pooling(conv_1, pooling_height, pooling_width_1st, pooling_stride)
# second CNN layer