def __init__(self, config):
self.cfg = config
self.target = []
self.inp = tf.placeholder(tf.float32, self.cfg.test_param_dims)
self.initialized = False
with tf.device('/gpu:0'):
with tf.variable_scope("model", reuse=tf.AUTO_REUSE) as scope:
self.model = cnn_model_struct()
self.model.build(self.inp, self.cfg.test_param_dims[1:], self.cfg.output_hist_dims[1:], train_mode=False, verbose=False)
self.gpuconfig = tf.ConfigProto()
self.gpuconfig.gpu_options.allow_growth = True
self.gpuconfig.allow_soft_placement = True
self.saver = tf.train.Saver()
def __init__(self, config, max_iters=100, tol=1e-7, nsamples=1e5):
self.max_iters = max_iters
self.tol = tol
self.N = nsamples
self.cfg = config
self.target = []
self.inf_batch_size = 50000
# placeholder for forward model
self.forward_model_inpdims = [self.inf_batch_size] + self.cfg.param_dims[1:]
self.forward_input = tf.placeholder(tf.float32, self.forward_model_inpdims)
self.forward_initialized = False
with tf.device('/gpu:0'):
with tf.variable_scope("model", reuse=tf.AUTO_REUSE) as scope:
# build the forward model
self.forward_model = cnn_model_struct()
self.forward_model.build(self.forward_input, self.cfg.param_dims[1:], self.cfg.output_hist_dims[1:], train_mode=False, verbose=False)
self.gpuconfig = tf.ConfigProto()
self.gpuconfig.gpu_options.allow_growth = True
self.gpuconfig.allow_soft_placement = True
self.saver = tf.train.Saver()
self.forward_sess = tf.Session(config=self.gpuconfig)
ckpts = tf.train.latest_checkpoint(self.cfg.model_output)
self.saver.restore(self.forward_sess, ckpts)
# placeholder for inverse model
self.inv_model_inpdims = [1] + self.cfg.output_hist_dims[1:]
self.inv_input = tf.placeholder(tf.float32, self.inv_model_inpdims)
with tf.device('/gpu:1'):
with tf.variable_scope("reversemodel", reuse=tf.AUTO_REUSE) as scope:
# build the inverse model
self.inv_model = cnn_reverse_model()
self.inv_model.build(self.inv_input, self.cfg.output_hist_dims[1:], self.cfg.param_dims[1:], train_mode=False, verbose=False)
self.gpuconfig1 = tf.ConfigProto()
self.gpuconfig1.gpu_options.allow_growth = True
self.gpuconfig1.allow_soft_placement = True
self.saver1 = tf.train.Saver(var_list=tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope='reversemodel'))
self.inv_sess = tf.Session(config=self.gpuconfig1)
ckpts = tf.train.latest_checkpoint(os.path.join(self.cfg.base_dir, 'models', 'rev_'+self.cfg.model_name+'_training_data_binned_{}_nbins_{}_n_{}'.format(int(self.cfg.isBinned),self.cfg.nBins,self.cfg.nDatapoints))
)
self.saver1.restore(self.inv_sess, ckpts)
def __init__(self,
config,
max_iters=100,
tol=1e-7,
nsamples=1e5,
tdist=False):
if tdist:
self.generateFromProposal = self.generateFromProposalTDist
self.evalProposalByComponent = self.evalProposalByComponentTDist
else:
self.generateFromProposal = self.generateFromProposalNormal
self.evalProposalByComponent = self.evalProposalByComponentNormal
self.max_iters = max_iters
self.tol = tol
self.N = nsamples
self.cfg = config
self.target = []
self.inf_batch_size = 50000
# placeholder for forward model
#self.forward_model_inpdims = [self.inf_batch_size] + self.cfg.param_dims[1:]
self.forward_model_inpdims = [None] + self.cfg.param_dims[1:]
self.forward_input = tf.placeholder(tf.float32,
self.forward_model_inpdims)
self.forward_initialized = False
with tf.device('/gpu:0'):
with tf.variable_scope("model", reuse=tf.AUTO_REUSE) as scope:
# build the forward model
self.forward_model = cnn_model_struct()
self.forward_model.build(self.forward_input,
self.cfg.param_dims[1:],
self.cfg.output_hist_dims[1:],
train_mode=False,
verbose=False)
self.gpuconfig = tf.ConfigProto()
self.gpuconfig.gpu_options.allow_growth = True
self.gpuconfig.allow_soft_placement = True
self.saver = tf.train.Saver()
self.forward_sess = tf.Session(config=self.gpuconfig)
ckpts = tf.train.latest_checkpoint(self.cfg.model_output)
self.saver.restore(self.forward_sess, ckpts)
'''
def __init__(self,config):
self.config = config
self.input = tf.placeholder(tf.float32,
[None,config.image_target_size[0],config.image_target_size[1],config.image_target_size[2]], name='ip_placeholder')
self.initialized = False
with tf.device('/gpu:0'):
with tf.variable_scope("model") as scope:
self.model = cnn_model_struct()
self.model.build(self.input, config.num_classes, train_mode=False)
self.gpuconfig = tf.ConfigProto()
self.gpuconfig.gpu_options.allow_growth = True
self.gpuconfig.allow_soft_placement = True
self.saver = tf.train.Saver()