def construct_model(obs_dim=11,
act_dim=3,
rew_dim=1,
hidden_dim=200,
num_networks=7,
num_elites=5,
session=None):
print(
'[ BNN ] Observation dim {} | Action dim: {} | Hidden dim: {}'.format(
obs_dim, act_dim, hidden_dim))
params = {
'name': 'BNN',
'num_networks': num_networks,
'num_elites': num_elites,
'sess': session
}
# num_networks是model ensemble的数量
model = BNN(params)
# 第一层必须指定input，后面层的input可以自动计算
model.add(
FC(hidden_dim,
input_dim=obs_dim + act_dim,
activation="swish",
weight_decay=0.000025))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.00005))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(obs_dim + rew_dim, weight_decay=0.0001))
model.finalize(tf.train.AdamOptimizer, {"learning_rate": 0.001})
return model
def construct_model(name='BNN',
obs_dim=11,
act_dim=3,
rew_dim=1,
hidden_dim=200,
num_networks=7,
num_elites=5,
q_func=None,
classifier=None,
use_classifier=False,
is_classifier=False,
session=None):
print(
'[ BNN ] Observation dim {} | Action dim: {} | Hidden dim: {}'.format(
obs_dim, act_dim, hidden_dim))
params = {
'name': name,
'num_networks': num_networks,
'num_elites': num_elites,
'q_func': q_func,
'classifier': classifier,
'use_classifier': use_classifier,
'is_classifier': is_classifier,
'sess': session
}
model = BNN(params)
if is_classifier:
model.add(
FC(hidden_dim,
input_dim=obs_dim * 2 + act_dim + rew_dim,
activation="swish",
weight_decay=0.000025))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.00005))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(1, activation="sigmoid", weight_decay=0.0001))
model.finalize(tf.train.AdamOptimizer, {"learning_rate": 0.001})
else:
model.add(
FC(hidden_dim,
input_dim=obs_dim + act_dim,
activation="swish",
weight_decay=0.000025))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.00005))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(obs_dim + rew_dim, weight_decay=0.0001))
model.finalize(tf.train.AdamOptimizer, {"learning_rate": 0.001})
return model
def _load_structure(self):
"""Uses the saved structure in self.model_dir with the name of this network to initialize
the structure of this network.
"""
structure = []
print("=========Inside load_sturcture:======== ", self.model_dir,
self.name)
with open(os.path.join(self.model_dir, "%s.nns" % self.name),
"r") as f:
for line in f:
kwargs = {
key: val
for (key, val) in
[argval.split("=") for argval in line[3:-2].split(", ")]
}
kwargs["input_dim"] = int(kwargs["input_dim"])
kwargs["output_dim"] = int(kwargs["output_dim"])
kwargs["weight_decay"] = None if kwargs[
"weight_decay"] == "None" else float(
kwargs["weight_decay"])
kwargs["activation"] = None if kwargs[
"activation"] == "None" else kwargs["activation"][1:-1]
kwargs["ensemble_size"] = int(kwargs["ensemble_size"])
structure.append(FC(**kwargs))
self.layers = structure
def construct_model(obs_dim=11,
act_dim=3,
rew_dim=1,
hidden_dim=200,
num_networks=7,
num_elites=5,
session=None,
model_dir=None,
model_load_timestep=None,
load_model=False):
print(
'[ BNN ] Observation dim {} | Action dim: {} | Hidden dim: {}'.format(
obs_dim, act_dim, hidden_dim))
name = 'BNN' if not model_load_timestep else 'BNN_' + str(
model_load_timestep)
params = {
'name': name,
'num_networks': num_networks,
'num_elites': num_elites,
'sess': session,
'model_dir': model_dir,
'load_model': load_model
}
model = BNN(params)
if not load_model:
model.add(
FC(hidden_dim,
input_dim=obs_dim + act_dim,
activation="swish",
weight_decay=0.000025))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.00005))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(obs_dim + rew_dim, weight_decay=0.0001))
model.finalize(tf.train.AdamOptimizer, {"learning_rate": 0.001})
return model
def construct_model(obs_dim=11,
act_dim=3,
rew_dim=1,
hidden_dim=200,
num_networks=7,
num_elites=5,
load_model=False,
model_dir=None,
session=None):
print(
'[ BNN ] Observation dim {} | Action dim: {} | Hidden dim: {}'.format(
obs_dim, act_dim, hidden_dim))
params = {
'name': 'BNN',
'num_networks': num_networks,
'num_elites': num_elites,
'sess': session
}
if load_model:
params['load_model'] = True
params['model_dir'] = model_dir
model = BNN(params)
if not load_model:
model.add(
FC(hidden_dim,
input_dim=obs_dim + act_dim,
activation="swish",
weight_decay=0.000025))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.00005))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(hidden_dim, activation="swish", weight_decay=0.000075))
model.add(FC(obs_dim + rew_dim, weight_decay=0.0001))
model.finalize(tf.train.AdamOptimizer, {"learning_rate": 0.001})
return model
class BNN:
"""Neural network models which model aleatoric uncertainty (and possibly epistemic uncertainty
with ensembling).
"""
def __init__(self, params):
"""Initializes a class instance.
Arguments:
params (DotMap): A dotmap of model parameters.
.name (str): Model name, used for logging/use in variable scopes.
Warning: Models with the same name will overwrite each other.
.num_networks (int): (optional) The number of networks in the ensemble. Defaults to 1.
Ignored if model is being loaded.
.model_dir (str/None): (optional) Path to directory from which model will be loaded, and
saved by default. Defaults to None.
.load_model (bool): (optional) If True, model will be loaded from the model directory,
assuming that the files are generated by a model of the same name. Defaults to False.
.sess (tf.Session/None): The session that this model will use.
If None, creates a session with its own associated graph. Defaults to None.
"""
self.name = get_required_argument(params, 'name', 'Must provide name.')
self.model_dir = params.get('model_dir', None)
print(
'[ BNN ] Initializing model: {} | {} networks | {} elites'.format(
params['name'], params['num_networks'], params['num_elites']))
if params.get('sess', None) is None:
config = tf.ConfigProto()
# config.gpu_options.allow_growth = True
self._sess = tf.Session(config=config)
else:
self._sess = params.get('sess')
# Instance variables
# self.action_group = params['action_group']
self.action_group = params['action_group']
self.obs_dimension = params['obs_dim']
self.action_dimension = params['act_dim']
self.group_numbers = len(self.action_group)
self.masks = self.get_masks()
self.finalized = False
self.layers, self.max_logvar, self.min_logvar = [
[] for _ in range(self.group_numbers)
], None, None
self.decays, self.optvars, self.nonoptvars = [], [], []
self.end_act, self.end_act_name = None, None
self.scaler = None
self.final_layer = None
# Training objects
self.optimizer = None
self.sy_train_in, self.sy_train_targ = None, None
self.train_op, self.mse_loss = None, None
# Prediction objects
self.sy_pred_in2d, self.sy_pred_mean2d_fac, self.sy_pred_var2d_fac = None, None, None
self.sy_pred_mean2d, self.sy_pred_var2d = None, None
self.sy_pred_in3d, self.sy_pred_mean3d_fac, self.sy_pred_var3d_fac = None, None, None
if params.get('load_model', False):
if self.model_dir is None:
raise ValueError(
"Cannot load model without providing model directory.")
self._load_structure()
self.num_nets, self.model_loaded = self.layers[0][
0].get_ensemble_size(), True
print("Model loaded from %s." % self.model_dir)
self.num_elites = params['num_elites']
else:
self.num_nets = params.get('num_networks', 1)
self.num_elites = params[
'num_elites'] #params.get('num_elites', 1)
self.model_loaded = False
if self.num_nets == 1:
print("Created a neural network with variance predictions.")
else:
print(
"Created an ensemble of {} neural networks with variance predictions | Elites: {}"
.format(self.num_nets, self.num_elites))
@property
def is_probabilistic(self):
return True
@property
def is_tf_model(self):
return True
@property
def sess(self):
return self._sess
def get_masks(self):
print("action:{}, state:{}".format(self.action_dimension,
self.obs_dimension))
return_list = []
mask_list_state = [1 for _ in range(self.obs_dimension)]
for i in self.action_group:
mask_list_action = [0 for _ in range(self.action_dimension)]
for j in i:
mask_list_action[j] = 1
mask_tensor = tf.constant(mask_list_state + mask_list_action,
dtype=tf.float32)
print(mask_list_action)
print(mask_tensor)
print(mask_tensor.shape)
return_list.append(mask_tensor)
return return_list
###################################
# Network Structure Setup Methods #
###################################
def add(self, layer):
"""Adds a new layer to the network.
Arguments:
layer: (layer) The new layer to be added to the network.
If this is the first layer, the input dimension of the layer must be set.
Returns: None.
"""
if self.finalized:
raise RuntimeError(
"Cannot modify network structure after finalizing.")
if len(self.layers[0]) == 0 and layer.get_input_dim() is None:
raise ValueError("Must set input dimension for the first layer.")
if self.model_loaded:
raise RuntimeError("Cannot add layers to a loaded model.")
layer.set_ensemble_size(self.num_nets)
if len(self.layers[0]) > 0:
layer.set_input_dim(self.layers[0][-1].get_output_dim())
for ed2_tmp in range(self.group_numbers):
self.layers[ed2_tmp].append(layer.copy())
def pop(self):
"""Removes and returns the most recently added layer to the network.
Returns: (layer) The removed layer.
"""
print("Focus pop")
assert 1 == 0
if len(self.layers[0]) == 0:
raise RuntimeError("Network is empty.")
if self.finalized:
raise RuntimeError(
"Cannot modify network structure after finalizing.")
if self.model_loaded:
raise RuntimeError("Cannot remove layers from a loaded model.")
return [
self.layers[ed2_tmp].pop() for ed2_tmp in range(self.group_numbers)
]
def finalize(self,
optimizer,
optimizer_args=None,
final_size=None,
*args,
**kwargs):
"""Finalizes the network.
Arguments:
optimizer: (tf.train.Optimizer) An optimizer class from those available at tf.train.Optimizer.
optimizer_args: (dict) A dictionary of arguments for the __init__ method of the chosen optimizer.
final_size: final layer size.
Returns: None
"""
if len(self.layers[0]) == 0:
raise RuntimeError("Cannot finalize an empty network.")
if self.finalized:
raise RuntimeError("Can only finalize a network once.")
optimizer_args = {} if optimizer_args is None else optimizer_args
self.optimizer = optimizer(**optimizer_args)
self.final_layer = FC(final_size, weight_decay=0.0001)
self.final_layer.set_ensemble_size(self.num_nets)
self.final_layer.set_input_dim(self.layers[0][-1].get_output_dim())
self.final_layer.set_output_dim(2 * self.final_layer.get_output_dim())
# Remove last activation to isolate variance from activation function.
self.end_act = self.final_layer.get_activation()
self.end_act_name = self.final_layer.get_activation(as_func=False)
self.final_layer.unset_activation()
# Construct all variables.
with self.sess.as_default():
with tf.variable_scope(self.name):
self.scaler = TensorStandardScaler(
self.layers[0][0].get_input_dim())
self.max_logvar = tf.Variable(
np.ones([1, self.final_layer.get_output_dim() // 2]) / 2.,
dtype=tf.float32,
name="max_log_var")
self.min_logvar = tf.Variable(
-np.ones([1, self.final_layer.get_output_dim() // 2]) *
10.,
dtype=tf.float32,
name="min_log_var")
for ed2_i, ed2 in enumerate(self.layers):
for i, layer in enumerate(ed2):
with tf.variable_scope("Layer{}{}".format(ed2_i, i)):
layer.construct_vars()
self.decays.extend(layer.get_decays())
self.optvars.extend(layer.get_vars())
self.final_layer.construct_vars()
self.decays.extend(self.final_layer.get_decays())
self.optvars.extend(self.final_layer.get_vars())
self.optvars.extend([self.max_logvar, self.min_logvar])
self.nonoptvars.extend(self.scaler.get_vars())
# Set up training
with tf.variable_scope(self.name):
self.expand_size_in3d = tf.placeholder(
dtype=tf.int32,
shape=[3],
name="3D_training_inputs_batch_size")
self.optimizer = optimizer(**optimizer_args)
self.sy_train_in = tf.placeholder(
dtype=tf.float32,
shape=[self.num_nets, None, self.layers[0][0].get_input_dim()],
name="training_inputs")
self.sy_train_targ = tf.placeholder(
dtype=tf.float32,
shape=[
self.num_nets, None,
self.final_layer.get_output_dim() // 2
],
name="training_targets")
train_loss = tf.reduce_sum(
self._compile_losses(self.sy_train_in,
self.expand_size_in3d,
self.sy_train_targ,
inc_var_loss=True))
train_loss += tf.add_n(self.decays)
train_loss += 0.01 * tf.reduce_sum(
self.max_logvar) - 0.01 * tf.reduce_sum(self.min_logvar)
self.mse_loss = self._compile_losses(self.sy_train_in,
self.expand_size_in3d,
self.sy_train_targ,
inc_var_loss=False)
self.train_op = self.optimizer.minimize(train_loss,
var_list=self.optvars)
# to here
# Initialize all variables
self.sess.run(
tf.variables_initializer(self.optvars + self.nonoptvars +
self.optimizer.variables()))
# Set up prediction
with tf.variable_scope(self.name):
self.expand_size_in2d = tf.placeholder(
dtype=tf.int32,
shape=[2],
name="2D_training_inputs_batch_size")
self.sy_pred_in2d = tf.placeholder(
dtype=tf.float32,
shape=[None, self.layers[0][0].get_input_dim()],
name="2D_training_inputs")
self.sy_pred_mean2d_fac, self.sy_pred_var2d_fac = \
self.create_prediction_tensors(self.sy_pred_in2d, self.expand_size_in2d, factored=True)
self.sy_pred_mean2d = tf.reduce_mean(self.sy_pred_mean2d_fac,
axis=0)
self.sy_pred_var2d = tf.reduce_mean(self.sy_pred_var2d_fac, axis=0) + \
tf.reduce_mean(tf.square(self.sy_pred_mean2d_fac - self.sy_pred_mean2d), axis=0)
self.sy_pred_in3d = tf.placeholder(
dtype=tf.float32,
shape=[self.num_nets, None, self.layers[0][0].get_input_dim()],
name="3D_training_inputs")
self.sy_pred_mean3d_fac, self.sy_pred_var3d_fac = \
self.create_prediction_tensors(self.sy_pred_in3d, self.expand_size_in3d, factored=True)
# Load model if needed
if self.model_loaded:
with self.sess.as_default():
params_dict = loadmat(
os.path.join(self.model_dir, "%s.mat" % self.name))
all_vars = self.nonoptvars + self.optvars
for i, var in enumerate(all_vars):
var.load(params_dict[str(i)])
self.finalized = True
##################
# Custom Methods #
##################
def _save_state(self, idx):
self._state[idx] = [[
layer.get_model_vars(idx, self.sess) for layer in part
] for part in self.layers]
self.final_state_[idx] = self.final_layer.get_model_vars(
idx, self.sess)
def _set_state(self):
keys = ['weights', 'biases']
ops = []
num_layers = len(self.layers[0])
for part in range(len(self.layers)):
for layer in range(num_layers):
# net_state = self._state[i]
params = {
key: np.stack([
self._state[net][part][layer][key]
for net in range(self.num_nets)
])
for key in keys
}
ops.extend(self.layers[part][layer].set_model_vars(params))
params = {
key: np.stack(
[self.final_state_[net][key] for net in range(self.num_nets)])
for key in keys
}
ops.extend(self.final_layer.set_model_vars(params))
self.sess.run(ops)
def _save_best(self, epoch, holdout_losses):
updated = False
for i in range(len(holdout_losses)):
current = holdout_losses[i]
_, best = self._snapshots[i]
improvement = (best - current) / best
if improvement > 0.01:
self._snapshots[i] = (epoch, current)
self._save_state(i)
updated = True
improvement = (best - current) / best
# print('epoch {} | updated {} | improvement: {:.4f} | best: {:.4f} | current: {:.4f}'.format(epoch, i, improvement, best, current))
if updated:
self._epochs_since_update = 0
else:
self._epochs_since_update += 1
if self._epochs_since_update > self._max_epochs_since_update:
# print('[ BNN ] Breaking at epoch {}: {} epochs since update ({} max)'.format(epoch, self._epochs_since_update, self._max_epochs_since_update))
return True
else:
return False
def _start_train(self):
self._state = {}
self.final_state_ = {}
self._snapshots = {i: (None, 1e10) for i in range(self.num_nets)}
self._epochs_since_update = 0
def _end_train(self, holdout_losses):
sorted_inds = np.argsort(holdout_losses)
self._model_inds = sorted_inds[:self.num_elites].tolist()
print('Using {} / {} models: {}'.format(self.num_elites, self.num_nets,
self._model_inds))
def random_inds(self, batch_size):
inds = np.random.choice(self._model_inds, size=batch_size)
return inds
def reset(self):
print('[ BNN ] Resetting model')
[[layer.reset(self.sess) for layer in part] for part in self.layers]
self.final_layer.reset(self.sess)
def validate(self, inputs, targets):
inputs = np.tile(inputs[None], [self.num_nets, 1, 1])
targets = np.tile(targets[None], [self.num_nets, 1, 1])
losses = self.sess.run(self.mse_loss,
feed_dict={
self.sy_train_in:
inputs,
self.sy_train_targ:
targets,
self.expand_size_in3d:
np.array(
[inputs.shape[0], inputs.shape[1], 1])
})
mean_elite_loss = np.sort(losses)[:self.num_elites].mean()
return mean_elite_loss
#################
# Model Methods #
#################
def train(self,
inputs,
targets,
batch_size=32,
max_epochs=None,
max_epochs_since_update=5,
hide_progress=False,
holdout_ratio=0.0,
max_logging=5000,
max_grad_updates=None,
timer=None,
max_t=None):
"""Trains/Continues network training
Arguments:
inputs (np.ndarray): Network inputs in the training dataset in rows.
targets (np.ndarray): Network target outputs in the training dataset in rows corresponding
to the rows in inputs.
batch_size (int): The minibatch size to be used for training.
epochs (int): Number of epochs (full network passes that will be done.
hide_progress (bool): If True, hides the progress bar shown at the beginning of training.
Returns: None
"""
self._max_epochs_since_update = max_epochs_since_update
self._start_train()
break_train = False
def shuffle_rows(arr):
idxs = np.argsort(np.random.uniform(size=arr.shape), axis=-1)
return arr[np.arange(arr.shape[0])[:, None], idxs]
# Split into training and holdout sets
num_holdout = min(int(inputs.shape[0] * holdout_ratio), max_logging)
permutation = np.random.permutation(inputs.shape[0])
inputs, holdout_inputs = inputs[permutation[num_holdout:]], inputs[
permutation[:num_holdout]]
targets, holdout_targets = targets[permutation[num_holdout:]], targets[
permutation[:num_holdout]]
holdout_inputs = np.tile(holdout_inputs[None], [self.num_nets, 1, 1])
holdout_targets = np.tile(holdout_targets[None], [self.num_nets, 1, 1])
print('[ BNN ] Training {} | Holdout: {}'.format(
inputs.shape, holdout_inputs.shape))
with self.sess.as_default():
self.scaler.fit(inputs)
idxs = np.random.randint(inputs.shape[0],
size=[self.num_nets, inputs.shape[0]])
if hide_progress:
progress = Silent()
else:
progress = Progress(max_epochs)
if max_epochs:
epoch_iter = range(max_epochs)
else:
epoch_iter = itertools.count()
# else:
# epoch_range = trange(epochs, unit="epoch(s)", desc="Network training")
t0 = time.time()
grad_updates = 0
for epoch in epoch_iter:
for batch_num in range(int(np.ceil(idxs.shape[-1] / batch_size))):
batch_idxs = idxs[:, batch_num * batch_size:(batch_num + 1) *
batch_size]
self.sess.run(self.train_op,
feed_dict={
self.sy_train_in:
inputs[batch_idxs],
self.sy_train_targ:
targets[batch_idxs],
self.expand_size_in3d:
np.array([
inputs[batch_idxs].shape[0],
inputs[batch_idxs].shape[1], 1
])
})
grad_updates += 1
idxs = shuffle_rows(idxs)
if not hide_progress:
if holdout_ratio < 1e-12:
losses = self.sess.run(
self.mse_loss,
feed_dict={
self.sy_train_in:
inputs[idxs[:, :max_logging]],
self.sy_train_targ:
targets[idxs[:, :max_logging]],
self.expand_size_in3d:
np.array([
inputs[idxs[:, :max_logging]].shape[0],
inputs[idxs[:, :max_logging]].shape[1], 1
])
})
named_losses = [['M{}'.format(i), losses[i]]
for i in range(len(losses))]
progress.set_description(named_losses)
else:
losses = self.sess.run(
self.mse_loss,
feed_dict={
self.sy_train_in:
inputs[idxs[:, :max_logging]],
self.sy_train_targ:
targets[idxs[:, :max_logging]],
self.expand_size_in3d:
np.array([
inputs[idxs[:, :max_logging]].shape[0],
inputs[idxs[:, :max_logging]].shape[1], 1
])
})
holdout_losses = self.sess.run(
self.mse_loss,
feed_dict={
self.sy_train_in:
holdout_inputs,
self.sy_train_targ:
holdout_targets,
self.expand_size_in3d:
np.array([
holdout_inputs.shape[0],
holdout_inputs.shape[1], 1
])
})
named_losses = [['M{}'.format(i), losses[i]]
for i in range(len(losses))]
named_holdout_losses = [[
'V{}'.format(i), holdout_losses[i]
] for i in range(len(holdout_losses))]
named_losses = named_losses + named_holdout_losses + [[
'T', time.time() - t0
]]
progress.set_description(named_losses)
break_train = self._save_best(epoch, holdout_losses)
progress.update()
t = time.time() - t0
if break_train or (max_grad_updates
and grad_updates > max_grad_updates):
break
if max_t and t > max_t:
descr = 'Breaking because of timeout: {}! (max: {})'.format(
t, max_t)
progress.append_description(descr)
# print('Breaking because of timeout: {}! | (max: {})\n'.format(t, max_t))
# time.sleep(5)
break
progress.stamp()
if timer: timer.stamp('bnn_train')
self._set_state()
if timer: timer.stamp('bnn_set_state')
holdout_losses = self.sess.run(
self.mse_loss,
feed_dict={
self.sy_train_in:
holdout_inputs,
self.sy_train_targ:
holdout_targets,
self.expand_size_in3d:
np.array([holdout_inputs.shape[0], holdout_inputs.shape[1], 1])
})
if timer: timer.stamp('bnn_holdout')
self._end_train(holdout_losses)
if timer: timer.stamp('bnn_end')
val_loss = (np.sort(holdout_losses)[:self.num_elites]).mean()
model_metrics = {'val_loss': val_loss}
print('[ BNN ] Holdout', np.sort(holdout_losses), model_metrics)
return OrderedDict(model_metrics)
# return np.sort(holdout_losses)[]
# pdb.set_trace()
def predict(self, inputs, factored=False, *args, **kwargs):
"""Returns the distribution predicted by the model for each input vector in inputs.
Behavior is affected by the dimensionality of inputs and factored as follows:
inputs is 2D, factored=True: Each row is treated as an input vector.
Returns a mean of shape [ensemble_size, batch_size, output_dim] and variance of shape
[ensemble_size, batch_size, output_dim], where N(mean[i, j, :], diag([i, j, :])) is the
predicted output distribution by the ith model in the ensemble on input vector j.
inputs is 2D, factored=False: Each row is treated as an input vector.
Returns a mean of shape [batch_size, output_dim] and variance of shape
[batch_size, output_dim], where aggregation is performed as described in the paper.
inputs is 3D, factored=True/False: Each row in the last dimension is treated as an input vector.
Returns a mean of shape [ensemble_size, batch_size, output_dim] and variance of sha
[ensemble_size, batch_size, output_dim], where N(mean[i, j, :], diag([i, j, :])) is the
predicted output distribution by the ith model in the ensemble on input vector [i, j].
Arguments:
inputs (np.ndarray): An array of input vectors in rows. See above for behavior.
factored (bool): See above for behavior.
"""
if len(inputs.shape) == 2:
if factored:
return self.sess.run(
[self.sy_pred_mean2d_fac, self.sy_pred_var2d_fac],
feed_dict={
self.sy_pred_in2d: inputs,
self.expand_size_in2d: np.array([inputs.shape[0], 1])
})
else:
return self.sess.run(
[self.sy_pred_mean2d, self.sy_pred_var2d],
feed_dict={
self.sy_pred_in2d: inputs,
self.expand_size_in2d: np.array([inputs.shape[0], 1])
})
else:
return self.sess.run(
[self.sy_pred_mean3d_fac, self.sy_pred_var3d_fac],
feed_dict={
self.sy_pred_in3d:
inputs,
self.expand_size_in3d:
np.array([inputs.shape[0], inputs.shape[1], 1])
})
def create_prediction_tensors(self,
inputs,
expand_dimension,
factored=False,
*args,
**kwargs):
"""See predict() above for documentation.
"""
factored_mean, factored_variance = self._compile_outputs(
inputs, expand_dimension)
if inputs.shape.ndims == 2 and not factored:
mean = tf.reduce_mean(factored_mean, axis=0)
variance = tf.reduce_mean(tf.square(factored_mean - mean), axis=0) + \
tf.reduce_mean(factored_variance, axis=0)
return mean, variance
return factored_mean, factored_variance
def save(self, savedir, timestep):
"""Saves all information required to recreate this model in two files in savedir
(or self.model_dir if savedir is None), one containing the model structuure and the other
containing all variables in the network.
savedir (str): (Optional) Path to which files will be saved. If not provided, self.model_dir
(the directory provided at initialization) will be used.
"""
if not self.finalized:
raise RuntimeError()
model_dir = self.model_dir if savedir is None else savedir
# Write structure to file
with open(
os.path.join(model_dir,
'{}_{}.nns'.format(self.name, timestep)),
"w+") as f:
for part in self.layers:
for layer in part:
f.write("%s\n" % repr(layer))
last_layer_copy = self.final_layer.copy()
last_layer_copy.set_activation(self.end_act_name)
last_layer_copy.set_output_dim(last_layer_copy.get_output_dim() //
2)
f.write("%s\n" % repr(last_layer_copy))
# Save network parameters (including scalers) in a .mat file
var_vals = {}
for i, var_val in enumerate(
self.sess.run(self.nonoptvars + self.optvars)):
var_vals[str(i)] = var_val
savemat(
os.path.join(model_dir, '{}_{}.mat'.format(self.name, timestep)),
var_vals)
def _load_structure(self):
"""Uses the saved structure in self.model_dir with the name of this network to initialize
the structure of this network.
"""
print("load structure")
assert 1 == 0
structure = []
with open(os.path.join(self.model_dir, "%s.nns" % self.name),
"r") as f:
for line in f:
kwargs = {
key: val
for (key, val) in
[argval.split("=") for argval in line[3:-2].split(", ")]
}
kwargs["input_dim"] = int(kwargs["input_dim"])
kwargs["output_dim"] = int(kwargs["output_dim"])
kwargs["weight_decay"] = None if kwargs[
"weight_decay"] == "None" else float(
kwargs["weight_decay"])
kwargs["activation"] = None if kwargs[
"activation"] == "None" else kwargs["activation"][1:-1]
kwargs["ensemble_size"] = int(kwargs["ensemble_size"])
structure.append(FC(**kwargs))
self.layers = structure
#######################
# Compilation methods #
#######################
def mask_action(self, cur_out, id, expand_dimension):
mask = self.masks[id]
shape_list = []
batch_ex = tf.reduce_prod(expand_dimension)
if expand_dimension.shape[0] == 2:
print("============{}==============".format(
expand_dimension.shape))
mask = tf.expand_dims(mask, 0)
elif expand_dimension.shape[0] == 3:
print("============{}==============".format(
expand_dimension.shape))
mask = tf.expand_dims(mask, 0)
mask = tf.expand_dims(mask, 0)
else:
print(expand_dimension.shape)
print("Bug!!!!!!!!!!!")
assert 1 == 0
print("Mask shape:{}, tile_shape:{}".format(mask.shape,
expand_dimension.shape))
mask = tf.tile(mask, expand_dimension)
print(mask)
# print("shapeshapeshapeshapeshapeshapeshapeshapeshapeshapeshapeshapeshapeshape", cur_out.shape)
return cur_out * mask
def _compile_outputs(self, inputs, expand_dimension, ret_log_var=False):
"""Compiles the output of the network at the given inputs.
If inputs is 2D, returns a 3D tensor where output[i] is the output of the ith network in the ensemble.
If inputs is 3D, returns a 3D tensor where output[i] is the output of the ith network on the ith input matrix.
Arguments:
inputs: (tf.Tensor) A tensor representing the inputs to the network
ret_log_var: (bool) If True, returns the log variance instead of the variance.
Returns: (tf.Tensors) The mean and variance/log variance predictions at inputs for each network
in the ensemble.
"""
dim_output = self.final_layer.get_output_dim()
cur_out = self.scaler.transform(inputs)
cur_out_part_list = []
for id, part in enumerate(self.layers):
cur_out_part = self.mask_action(cur_out, id, expand_dimension)
for layer in part:
cur_out_part = layer.compute_output_tensor(cur_out_part)
cur_out_part_list.append(cur_out_part)
final_cur_out = tf.reduce_mean(tf.stack(cur_out_part_list), 0)
cur_out = self.final_layer.compute_output_tensor(final_cur_out)
mean = cur_out[:, :, :dim_output // 2]
if self.end_act is not None:
mean = self.end_act(mean)
logvar = self.max_logvar - tf.nn.softplus(
self.max_logvar - cur_out[:, :, dim_output // 2:])
logvar = self.min_logvar + tf.nn.softplus(logvar - self.min_logvar)
if ret_log_var:
return mean, logvar
else:
return mean, tf.exp(logvar)
def _compile_losses(self, inputs, expand_size, targets, inc_var_loss=True):
"""Helper method for compiling the loss function.
The loss function is obtained from the log likelihood, assuming that the output
distribution is Gaussian, with both mean and (diagonal) covariance matrix being determined
by network outputs.
Arguments:
inputs: (tf.Tensor) A tensor representing the input batch
targets: (tf.Tensor) The desired targets for each input vector in inputs.
inc_var_loss: (bool) If True, includes log variance loss.
Returns: (tf.Tensor) A tensor representing the loss on the input arguments.
"""
mean, log_var = self._compile_outputs(
inputs,
expand_size,
ret_log_var=True,
)
inv_var = tf.exp(-log_var)
if inc_var_loss:
mse_losses = tf.reduce_mean(tf.reduce_mean(
tf.square(mean - targets) * inv_var, axis=-1),
axis=-1)
var_losses = tf.reduce_mean(tf.reduce_mean(log_var, axis=-1),
axis=-1)
total_losses = mse_losses + var_losses
else:
total_losses = tf.reduce_mean(tf.reduce_mean(tf.square(mean -
targets),
axis=-1),
axis=-1)
return total_losses
def finalize(self,
optimizer,
optimizer_args=None,
final_size=None,
*args,
**kwargs):
"""Finalizes the network.
Arguments:
optimizer: (tf.train.Optimizer) An optimizer class from those available at tf.train.Optimizer.
optimizer_args: (dict) A dictionary of arguments for the __init__ method of the chosen optimizer.
final_size: final layer size.
Returns: None
"""
if len(self.layers[0]) == 0:
raise RuntimeError("Cannot finalize an empty network.")
if self.finalized:
raise RuntimeError("Can only finalize a network once.")
optimizer_args = {} if optimizer_args is None else optimizer_args
self.optimizer = optimizer(**optimizer_args)
self.final_layer = FC(final_size, weight_decay=0.0001)
self.final_layer.set_ensemble_size(self.num_nets)
self.final_layer.set_input_dim(self.layers[0][-1].get_output_dim())
self.final_layer.set_output_dim(2 * self.final_layer.get_output_dim())
# Remove last activation to isolate variance from activation function.
self.end_act = self.final_layer.get_activation()
self.end_act_name = self.final_layer.get_activation(as_func=False)
self.final_layer.unset_activation()
# Construct all variables.
with self.sess.as_default():
with tf.variable_scope(self.name):
self.scaler = TensorStandardScaler(
self.layers[0][0].get_input_dim())
self.max_logvar = tf.Variable(
np.ones([1, self.final_layer.get_output_dim() // 2]) / 2.,
dtype=tf.float32,
name="max_log_var")
self.min_logvar = tf.Variable(
-np.ones([1, self.final_layer.get_output_dim() // 2]) *
10.,
dtype=tf.float32,
name="min_log_var")
for ed2_i, ed2 in enumerate(self.layers):
for i, layer in enumerate(ed2):
with tf.variable_scope("Layer{}{}".format(ed2_i, i)):
layer.construct_vars()
self.decays.extend(layer.get_decays())
self.optvars.extend(layer.get_vars())
self.final_layer.construct_vars()
self.decays.extend(self.final_layer.get_decays())
self.optvars.extend(self.final_layer.get_vars())
self.optvars.extend([self.max_logvar, self.min_logvar])
self.nonoptvars.extend(self.scaler.get_vars())
# Set up training
with tf.variable_scope(self.name):
self.expand_size_in3d = tf.placeholder(
dtype=tf.int32,
shape=[3],
name="3D_training_inputs_batch_size")
self.optimizer = optimizer(**optimizer_args)
self.sy_train_in = tf.placeholder(
dtype=tf.float32,
shape=[self.num_nets, None, self.layers[0][0].get_input_dim()],
name="training_inputs")
self.sy_train_targ = tf.placeholder(
dtype=tf.float32,
shape=[
self.num_nets, None,
self.final_layer.get_output_dim() // 2
],
name="training_targets")
train_loss = tf.reduce_sum(
self._compile_losses(self.sy_train_in,
self.expand_size_in3d,
self.sy_train_targ,
inc_var_loss=True))
train_loss += tf.add_n(self.decays)
train_loss += 0.01 * tf.reduce_sum(
self.max_logvar) - 0.01 * tf.reduce_sum(self.min_logvar)
self.mse_loss = self._compile_losses(self.sy_train_in,
self.expand_size_in3d,
self.sy_train_targ,
inc_var_loss=False)
self.train_op = self.optimizer.minimize(train_loss,
var_list=self.optvars)
# to here
# Initialize all variables
self.sess.run(
tf.variables_initializer(self.optvars + self.nonoptvars +
self.optimizer.variables()))
# Set up prediction
with tf.variable_scope(self.name):
self.expand_size_in2d = tf.placeholder(
dtype=tf.int32,
shape=[2],
name="2D_training_inputs_batch_size")
self.sy_pred_in2d = tf.placeholder(
dtype=tf.float32,
shape=[None, self.layers[0][0].get_input_dim()],
name="2D_training_inputs")
self.sy_pred_mean2d_fac, self.sy_pred_var2d_fac = \
self.create_prediction_tensors(self.sy_pred_in2d, self.expand_size_in2d, factored=True)
self.sy_pred_mean2d = tf.reduce_mean(self.sy_pred_mean2d_fac,
axis=0)
self.sy_pred_var2d = tf.reduce_mean(self.sy_pred_var2d_fac, axis=0) + \
tf.reduce_mean(tf.square(self.sy_pred_mean2d_fac - self.sy_pred_mean2d), axis=0)
self.sy_pred_in3d = tf.placeholder(
dtype=tf.float32,
shape=[self.num_nets, None, self.layers[0][0].get_input_dim()],
name="3D_training_inputs")
self.sy_pred_mean3d_fac, self.sy_pred_var3d_fac = \
self.create_prediction_tensors(self.sy_pred_in3d, self.expand_size_in3d, factored=True)
# Load model if needed
if self.model_loaded:
with self.sess.as_default():
params_dict = loadmat(
os.path.join(self.model_dir, "%s.mat" % self.name))
all_vars = self.nonoptvars + self.optvars
for i, var in enumerate(all_vars):
var.load(params_dict[str(i)])
self.finalized = True
def construct_model(in_dim,
out_dim,
name='BNN',
hidden_dims=(200, 200, 200),
num_networks=7,
num_elites=5,
loss = 'NLL',
activation = 'swish',
output_activation = None,
decay=1e-4,
lr = 1e-3,
lr_decay = None,
decay_steps=None,
weighted=False,
use_scaler_in = False,
use_scaler_out = False,
sc_factor = 1,
clip_loss = False,
kl_cliprange = 0.1,
var_corr = False,
max_logvar = .5,
min_logvar = -6,
logit_bias_std=1.0,
session=None):
"""
Constructs a tf model.
Args:
loss: Choose from 'NLL', 'MSE', 'Huber', or 'CE'.
choosing NLL will construct a model with variance output
"""
print('[ BNN ] dim in / out: {} / {} | Hidden dim: {}'.format(in_dim, out_dim, hidden_dims))
#print('[ BNN ] Input Layer dim: {} | Output Layer dim: {} '.format(obs_dim_in+act_dim+prior_dim, obs_dim_out+rew_dim))
params = {'name': name,
'loss':loss,
'num_networks': num_networks,
'num_elites': num_elites,
'sess': session,
'use_scaler_in': use_scaler_in,
'use_scaler_out': use_scaler_out,
'sc_factor': sc_factor,
'clip_loss': clip_loss,
'kl_cliprange':kl_cliprange,
'var_corr': var_corr,
'max_logvar':max_logvar,
'min_logvar':min_logvar,
'logit_bias_std':logit_bias_std,
}
model = BNN(params)
model.add(FC(hidden_dims[0], input_dim=in_dim, activation=activation, weight_decay=decay/4)) # def dec: 0.000025))
for hidden_dim in hidden_dims[1:]:
model.add(FC(hidden_dim, activation=activation, weight_decay=decay/2)) # def dec: 0.00005))
model.add(FC(out_dim, activation=output_activation, weight_decay=decay)) # def dec: 0.0001
opt_params = {"learning_rate":lr} if lr_decay is None else {"learning_rate":lr,
"learning_rate_decay":lr_decay,
"decay_steps":decay_steps}
model.finalize(tf.train.AdamOptimizer, opt_params, weighted=weighted)
total_parameters = 0
for variable in tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=name):
# shape is an array of tf.Dimension
shape = variable.get_shape()
variable_parameters = 1
for dim in shape:
variable_parameters *= dim.value
total_parameters += variable_parameters
print('[ BNN ] Total trainable Parameteres: {} '.format(total_parameters))
return model