'Neural network structure: detection {} and modification {}'.format(
pargs.dnet, pargs.mnet))
dlayers = list(map(int, pargs.dnet.split('-')))
mlayers = list(map(int, pargs.mnet.split('-')))
dact = list(map(tf_network.parse_act, pargs.dact.split('-')))
mact = list(map(tf_network.parse_act, pargs.mact.split('-')))
init_w = {'w': {'stddev': 0.1}, 'b': {'value': 0.5}}
x = tf.placeholder(tf.float32, shape=[None, dlayers[0]], name='x-input')
keep_prob = tf.placeholder(
tf.float32, name='keep_prob_detector') if pargs.ddrop > 0.0 else None
keep_prob2 = tf.placeholder(
tf.float32, name='keep_prob_modifier') if pargs.mdrop > 0.0 else None
y_, y = tf_network.build_network(dlayers,
dact,
init_weights=init_w,
input_x_holder=x,
dropout_holder=keep_prob,
bn=pargs.bn,
scope='detector/')[1:]
y2_, y2 = tf_network.build_network(mlayers,
mact,
init_weights=init_w,
input_x_holder=x,
dropout_holder=keep_prob2,
bn=pargs.bn,
scope='modifier/')[1:]
if pargs.mve:
s = tf_network.build_network(mlayers,
mact,
init_weights=init_w,
input_x_holder=x,
import numpy as np
dtype_real = np.float32 # NOTE: if double precision, use float64
dtype_int = np.int32 # NOTE: if int in C is 64bits, use int64
np.random.seed(1)
import tensorflow as tf
tf_sess = tf.InteractiveSession()
import tf_network
dlayers = list(map(int, tfopt['dnet'].split('-')))
mlayers = list(map(int, tfopt['mnet'].split('-')))
dact = list(map(tf_network.parse_act, tfopt['dact'].split('-')))
mact = list(map(tf_network.parse_act, tfopt['mact'].split('-')))
x = tf.placeholder(tf.float32, shape=[None, dlayers[0]], name='x-input')
y_, y = tf_network.build_network(dlayers, dact, input_x_holder=x, bn=tfopt['bn'], is_training=False, scope='detector/')[1:]
y2_, y2 = tf_network.build_network(mlayers, mact, input_x_holder=x, bn=tfopt['bn'], is_training=False, scope='modifier/')[1:]
if tfopt['mve']:
sd = tf_network.build_network(mlayers, mact, input_x_holder=x, input_y_holder=y2_, bn=tfopt['bn'], is_training=False, scope='modifier_var/')[2]
tf_saver = tf.train.Saver()
modelfile = pargs.load + '/model.ckpt'
tf_saver.restore(tf_sess, modelfile)
print('Pre-trained model {} loaded\n'.format(modelfile))
import manta as mt
mt.tFluid = FlagFluid
mt.tObstacle = FlagObstacle
nogui = False
pause = False