true_br = br_pairs[0]
true_wf = wf_pairs[0]
waveform_max = int(true_wf.size / waveform_reduction_factor)
# true_wf = true_wf[::waveform_reduction_factor]
true_wf = true_wf[:waveform_max]
# reshape for mono waveforms
true_wf = true_wf.reshape((-1, 1))
# ################
# MODEL DEFINITION
# ################
bits_per_second = true_wf.size / 10
train_flag, x, model = deep_residual_network(true_wf.dtype,
true_wf.shape,
tensorboard_output=False)
# placeholder for the truth label
y_true = tf.placeholder(true_wf.dtype, shape=x.get_shape())
# ################
# ################
# #############
# LOSS FUNCTION
# #############
with tf.name_scope('waveform_mse'):
waveform_mse = tf.reduce_mean(tf.square(tf.subtract(y_true, model)))
tf.summary.scalar('waveform_mse', waveform_mse)
if END_OFFSET == 0:
true_wf = true_wf[BEGIN_OFFSET * true_br:]
ds_wf = ds_wf[BEGIN_OFFSET * true_br:]
else:
true_wf = true_wf[BEGIN_OFFSET * true_br:END_OFFSET * true_br]
ds_wf = ds_wf[BEGIN_OFFSET * true_br:END_OFFSET * true_br]
true_wf = true_wf[:int(true_wf.size / INPUT_SIZE) * INPUT_SIZE]
ds_wf = ds_wf[:int(ds_wf.size / INPUT_SIZE) * INPUT_SIZE]
number_of_reco_iterations = int(ds_wf.size / INPUT_SIZE)
# ################
# MODEL DEFINITION
# ################
train_flag, x, model = deep_residual_network(true_wf.dtype,
true_wf.shape,
**model_settings)
# ################
# ################
# Add ops to restore all the variables.
saver = tf.train.Saver()
# create session
sess = tf.Session()
# restore model from checkpoint file
saver.restore(sess, model_checkpoint_file_name)