def validated_tf_function(x, internals, return_internals=False):
if util.is_atomic_values_spec(values_spec=self.inputs_spec):
if not util.is_consistent_with_value_spec(value_spec=self.inputs_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
else:
if not all(
util.is_consistent_with_value_spec(value_spec=spec, x=x[name])
for name, spec in self.inputs_spec.items()
):
raise TensorforceError("Invalid input arguments for tf_apply.")
if not all(
util.is_consistent_with_value_spec(value_spec=spec, x=internals[name])
for name, spec in self.__class__.internals_spec(network=self).items()
):
raise TensorforceError("Invalid input arguments for tf_apply.")
if return_internals:
x, internals = tf_function(x=x, internals=internals, return_internals=True)
else:
x = tf_function(x=x, internals=internals, return_internals=False)
if not util.is_consistent_with_value_spec(value_spec=self.get_output_spec(), x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
if return_internals and not all(
util.is_consistent_with_value_spec(value_spec=spec, x=internals[name])
for name, spec in self.__class__.internals_spec(network=self).items()
):
raise TensorforceError("Invalid output arguments for tf_apply.")
if return_internals:
return x, internals
else:
return x
def validated_tf_function(xs):
x = xs[:, 0, :]
if not util.is_consistent_with_value_spec(value_spec=self.input_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
x = tf_function(xs=xs)
if not util.is_consistent_with_value_spec(value_spec=self.output_spec, x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
return x
def validated_tf_function(x, previous):
if not util.is_consistent_with_value_spec(value_spec=self.input_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
# previous spec!
x, previous = tf_function(x=x, previous=previous)
if not util.is_consistent_with_value_spec(value_spec=self.output_spec, x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
return x, previous
def validated_tf_function(x):
if self.input_spec is not None and \
not util.is_consistent_with_value_spec(value_spec=self.input_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
x = tf_function(x=x)
if self.output_spec is not None and \
not util.is_consistent_with_value_spec(value_spec=self.output_spec, x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
return x
def validated_tf_function(x):
if self.input_spec is not None and \
not util.is_consistent_with_value_spec(value_spec=self.input_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='input',
value=x)
x = tf_function(x=x)
if self.output_spec is not None and \
not util.is_consistent_with_value_spec(value_spec=self.output_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='output',
value=x)
return x
def update_tensor(name, tensor):
# for n in range(len(Module.global_scope) + 1):
# partial_scope = Module.global_scope[:len(Module.global_scope) - n]
# scoped_name = util.join_scopes(*partial_scope, name)
# if scoped_name in Module.global_tensors_spec:
# break
# else:
# raise TensorforceError("Global tensor is not registered: {}.".format(name))
if name not in Module.global_tensors_spec:
raise TensorforceError(
"Global tensor is not registered: {}.".format(name))
scoped_name = name
spec = Module.global_tensors_spec[scoped_name]
if not util.is_consistent_with_value_spec(value_spec=spec, x=tensor):
raise TensorforceError(
"Invalid overwriting tensor: {}.".format(tensor))
scoped_name = util.join_scopes(*Module.global_scope, name)
previous = Module.global_tensors.get(scoped_name)
Module.global_tensors[scoped_name] = tensor
return previous
def validated_tf_function(xs):
x = xs[:, 0, :]
if not util.is_consistent_with_value_spec(
value_spec=self.input_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='input',
value=x)
x = tf_function(xs=xs)
if not util.is_consistent_with_value_spec(
value_spec=self.output_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='output',
value=x)
return x
def validated_tf_function(x):
if util.is_atomic_values_spec(values_spec=self.inputs_spec):
if not util.is_consistent_with_value_spec(value_spec=self.inputs_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
else:
if not all(
util.is_consistent_with_value_spec(value_spec=spec, x=x[name])
for name, spec in self.inputs_spec.items()
):
raise TensorforceError("Invalid input arguments for tf_apply.")
x = tf_function(x=x)
if not util.is_consistent_with_value_spec(value_spec=self.get_output_spec(), x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
return x
def validated_tf_function(x, previous):
if not util.is_consistent_with_value_spec(
value_spec=self.input_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='input',
value=x)
# previous spec!
x, previous = tf_function(x=x, previous=previous)
if not util.is_consistent_with_value_spec(
value_spec=self.output_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='output',
value=x)
return x, previous
def validated_tf_function(x, initial=None):
if not util.is_consistent_with_value_spec(value_spec=self.input_spec, x=x):
raise TensorforceError("Invalid input arguments for tf_apply.")
# initial spec!
if initial is None:
x = tf_function(x=x)
else:
x, final = tf_function(x=x, initial=initial)
if not util.is_consistent_with_value_spec(value_spec=self.output_spec, x=x):
raise TensorforceError("Invalid output arguments for tf_apply.")
if initial is None:
return x
else:
return x, final
def validated_tf_function(x, initial=None):
if not util.is_consistent_with_value_spec(
value_spec=self.input_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='input',
value=x)
# initial spec!
if initial is None:
x = tf_function(x=x)
else:
x, final = tf_function(x=x, initial=initial)
if not util.is_consistent_with_value_spec(
value_spec=self.output_spec, x=x):
raise TensorforceError.value(name='layer.apply',
argument='output',
value=x)
if initial is None:
return x
else:
return x, final
def update_tensor(name, tensor):
if name not in Module.global_tensors_spec:
raise TensorforceError("Global tensor is not registered: {}.".format(name))
scoped_name = name
spec = Module.global_tensors_spec[scoped_name]
if not util.is_consistent_with_value_spec(value_spec=spec, x=tensor):
raise TensorforceError("Invalid overwriting tensor: {}.".format(tensor))
scoped_name = util.join_scopes(*Module.global_scope, name)
previous = Module.global_tensors.get(scoped_name)
Module.global_tensors[scoped_name] = tensor
if Module.cond_counter == 0 and Module.while_counter == 0:
Module.queryable_tensors[scoped_name] = tensor
return previous
def update_tensor(name, tensor):
if name not in Module.global_tensors_spec:
raise TensorforceError.value(
name='Module.update_tensor', argument='name', value=name
)
scoped_name = name
spec = Module.global_tensors_spec[scoped_name]
if not util.is_consistent_with_value_spec(value_spec=spec, x=tensor):
raise TensorforceError.value(
name='Module.update_tensor', argument='tensor', value=tensor
)
scoped_name = util.join_scopes(*Module.global_scope, name)
previous = Module.global_tensors.get(scoped_name)
Module.global_tensors[scoped_name] = tensor
if Module.cond_counter == 0 and Module.while_counter == 0:
Module.queryable_tensors[scoped_name] = tensor
return previous
