def _get_params(self,
floating: Optional[bool] = True,
is_yield: Optional[bool] = None,
extract_independent: Optional[bool] = True) -> Set["ZfitParameter"]:
params = OrderedSet()
params = params.union(*(model.get_params(floating=floating, is_yield=is_yield,
extract_independent=extract_independent)
for model in self.model))
params = params.union(*(constraint.get_params(floating=floating, is_yield=False,
extract_independent=extract_independent)
for constraint in self.constraints))
return params
def _order(self, cleaned):
# TODO: Word == (adjective or noun) && (plural) && (before a noun) -> make singular e.g.:
# - Five easy bananas minutes. (Weird-as-a-Service)
# - Five easy banana minutes. (Totally sensible Driven Development)
mapped = {}
for words in cleaned:
for key, value in words.items():
for kind in self._ORDERING:
if kind in value.get('categories'):
mapped.setdefault(kind, []).append(key)
remove_empty = list(
filter(None, [
mapped.get('number'),
mapped.get('adverb'),
mapped.get('adjective'),
mapped.get('noun'),
]))
# Order the words in the sentence using a basic English language syntax.
cartesian_product = list(product(*remove_empty))
sentences = [
list(self._flat_tuple(item)) for item in cartesian_product
]
duplicates_removed = OrderedSet(
[tuple(OrderedSet(sentence)) for sentence in sentences])
# Having no numbers at the start of otherwise valid sentences is legit English.
no_need_for_numbers = OrderedSet([
tuple(self._get_rest(sentence)) for sentence in duplicates_removed
if isinstance(self._get_first(sentence), int) and len(sentence) > 1
])
return duplicates_removed.union(no_need_for_numbers)
def get_clique_category(
clique_graph: nx.MultiDiGraph, clique: List
) -> Tuple[str, List]:
"""
Given a clique, identify the category of the clique.
Parameters
----------
clique_graph: nx.MultiDiGraph
Clique graph
clique: List
A list of nodes in clique
Returns
-------
Tuple[str, list]
A tuple of clique category and its ancestors
"""
l = [clique_graph.nodes()[x]["category"] for x in clique]
u = OrderedSet.union(*l)
uo = sort_categories(u)
log.debug(f"outcome of union (sorted): {uo}")
clique_category = uo[0]
clique_category_ancestors = get_biolink_ancestors(uo[0])
return clique_category, clique_category_ancestors
def header_elements(self):
"""
:returns: The set of `css` dependencies for all page elements
in the page
"""
elements = OrderedSet()
elements = elements.union(*(content.header_elements
for content in self.contents))
return elements
def universal_set(documents, k) :
universal = OrderedSet()
universal1 = OrderedSet()
for doc in documents :
#print("doc = ", doc)
shingles, shings = k_shingles(doc, k)
universal = universal.union(shingles)
universal1 = universal1.union(shings)
universal = list(universal)
universal1 = list(universal1)
universal.sort()
universal1.sort()
universal = OrderedSet(universal)
universal1 = OrderedSet(universal1)
return universal, universal1
def _get_params(
self,
floating: bool | None = True,
is_yield: bool | None = None,
extract_independent: bool | None = True,
) -> set[ZfitParameter]:
params = OrderedSet()
params = params.union(*(model.get_params(
floating=floating,
is_yield=is_yield,
extract_independent=extract_independent,
) for model in self.model))
params = params.union(*(constraint.get_params(
floating=floating,
is_yield=False,
extract_independent=extract_independent,
) for constraint in self.constraints))
return params
def extract_complete_file_info(graph: RevisionGraph, fn_get_output):
'''
Given a revision graph, collect all file info
using tokei for those revisions
:param: graph - Revision Graph
:return: None. As a side effect, compliete file info by language type will be added to all
revisions in master_rev
'''
# We use a custom .mailmap file to resolve autors. That was great while
# we collected revision info (extract_revision_graph), but will block our ability to
# checkout individual revisions, therefor, clear the .mailmap checkout (if it exists)
fn_get_output(['git checkout -- .mailmap >/dev/null 2>&1'])
collect_file_info(graph.revisions.values(), fn_get_output)
fn_get_output(['git checkout master >/dev/null 2>&1'])
collect_deltas(graph, OrderedSet.union(graph.master_revs,
graph.not_a_merge))
def finalise(self):
return OrderedSet.union(*self.vocabs)
def _check_convert_input(self, loss: ZfitLoss, params, init=None, floating=True
) -> Tuple[ZfitLoss, Iterable[ZfitParameter], Union[None, FitResult]]:
"""Sanitize the input values and return all of them.
Args:
loss: If the loss is a callable, it will be converted to a SimpleLoss.
params: If the parameters is an array, it will be converted to free parameters.
init:
floating:
Returns:
loss, params, init:
"""
if isinstance(loss, ZfitResult):
init = loss # make the names correct
loss = init.loss
if params is None:
params = list(init.params)
to_set_param_values = {}
if isinstance(params, collections.Mapping) and all(isinstance(p, ZfitParameter) for p in params):
to_set_param_values = {p: val for p, val in params.items() if val is not None}
params = list(params.keys())
# convert the function to a SimpleLoss
if not isinstance(loss, ZfitLoss):
if not callable(loss):
raise TypeError("Given Loss has to be a ZfitLoss or a callable.")
elif params is None:
raise ValueError("If the loss is a callable, the params cannot be None.")
from zfit.core.loss import SimpleLoss
params = convert_to_parameters(params, prefer_constant=False)
loss = SimpleLoss(func=loss, params=params)
if params is None:
params = loss.get_params(floating=floating)
else:
if to_set_param_values:
try:
assign_values(list(to_set_param_values), list(to_set_param_values.values()))
except ParameterNotIndependentError as error:
not_indep_and_set = {p for p, val in to_set_param_values.items()
if val is not None and not p.independent}
raise ParameterNotIndependentError(f"Cannot set parameter {not_indep_and_set} to a value as they"
f" are not independent. The following `param` argument was"
f" given: {params}."
f""
f"Original error"
f"--------------"
f"{error}") from error
else:
params = convert_to_container(params, container=OrderedSet)
# now extract all the independent parameters
params = list(OrderedSet.union(*(p.get_params(only_floating=floating) for p in params)))
# set the parameter values from the init
if init is not None:
# don't set the user set
params_to_set = OrderedSet(params).intersection(OrderedSet(init.params)) - OrderedSet(to_set_param_values)
assign_values(params_to_set, init)
if floating:
params = self._filter_floating_params(params)
if not params:
raise RuntimeError("No parameter for minimization given/found. Cannot minimize.")
params = list(params)
return loss, params, init
class GSet(StateCRDT, OrderedSet):
def __init__(self, iterable=None, options=None):
self._payload = OrderedSet() if iterable is None else OrderedSet(
iterable)
# self._operations = [] if iterable is None \
# else map(OperationTuple3.create,iterable)
self._options = {} if options is None else options
def merge(self, other):
assert isinstance(other, GSet)
# print("Pre merge:")
# print(self._payload)
# print(other._payload)
merged = list(self._payload.union(other._payload))
# print("******************* PERFORMING MERGE ***************")
# print(merged)
merged.sort()
# print("--")
# print(merged)
return GSet(merged)
def compare(self, other):
return self.issubset(other)
# @property
def values(self):
return self._payload.__iter__()
def get_payload(self):
return list(self._payload)
def set_payload(self, payload):
self._payload = OrderedSet(payload)
def generate_log(self, log):
assert isinstance(log, GSet)
return GSet(log.get_payload())
payload = property(get_payload, set_payload)
#
# Set API
#
def add(self, element):
self._payload.add(element)
temp = list(self._payload)
temp.sort()
self._payload = OrderedSet(temp)
def discard(self, element):
raise NotImplementedError("This is a grow-only set")
def __contains__(self, element):
return self.values.__contains__(element)
def __iter__(self):
return self.values.__iter__()
def __len__(self):
return self._payload.__len__()
def call(self, X, adjM, parts):
#print("ok")
# extract number of neurons from adjM number of rows (adjM is 2D square matrix)
# this is here for option to decrease number of neurons in the following layers by shrinking adjM
# e.g. neurons over leaf nodes in graph
num_neurons = len(adjM)
# contains information which neurons to gather signal from (for every neuron list)
receptive_fields = [
tf.where(adjM[i] == 1)[:, 0] for i in range(num_neurons)
]
# if self.q_dim_security == 'adjM':
# newAdjM = np.array(adjM)
# indices = list(zip(*np.where(adjM == 1)))
# for row,col in indices:
# newAdjM[row] += adjM[col]
# oldAdjM = adjM
# adjM=np.clip(newAdjM, a_min=0, a_max=1)
#
# new_parts_delta = [
# OrderedSet(np.where(oldAdjM[i] == 1)[0].tolist())
# for i in range(num_neurons)]
# new_parts = [OrderedSet.union(parts[i], new_parts_delta[i]) for i in range(num_neurons)]
# else:
# new, cumulative receptive fields (parts) based on adjM (for every neuron in current layer)
# for every neuron i;
# parts of every neuron in the receptive field of 'i' are reduced with union to get cumulative receptive fields
# we iter through parts so that initial order of parts is preserved (because it is left operand of union)
# we do only if it is part of receptive field so that original logic is kept
# also we use union with receptive field to make sure the first step has receptive field inside parts (further unions wont change a thing)
new_parts = [
reduce(OrderedSet.union, [
parts[neighbour_idx] for neighbour_idx in OrderedSet.union(
parts[node_idx], receptive_fields[node_idx].numpy().tolist(
)) if neighbour_idx in receptive_fields[node_idx]
]) for node_idx in range(num_neurons)
]
# for every neuron i;
# create promotion chi matrix for every neuron/node in i's receptive field
chis = [{
neighbour_idx.numpy():
tf.convert_to_tensor(self.permutationFunction(
parts[neighbour_idx], new_parts[i]),
dtype=tf.float32)
for neighbour_idx in receptive_fields[i]
} for i in range(num_neurons)]
# for every neuron i;
# promote every activation of nodes in i's receptive field
# IMPORTANT:
# (probably) This is where tf functions should start to be used because new structures are formed based on previous ones
# and these new structures will ultimately 'transform' and mix with W to create activations
# we use new_parts for neighbour_idx (even there might be no promotion) so that we can gather promotions
# and fill 0's for next step in appropriate order
promotions = [[
tf.einsum(
self.einsum_expr,
*([chis[i][neighbour_idx]] * self.k +
[X[neighbour_idx]])) if neighbour_idx in receptive_fields[i]
else tf.zeros([self.channels_in] + [len(new_parts[i])] * self.k +
self.feature_vector_shape)
for neighbour_idx in new_parts[i]
] for i in range(num_neurons)]
##
stacked = [tf.stack(promotions[i], axis=1) for i in range(num_neurons)]
if self.mix_promotions_with_adjM:
adjMs_forNodes = [
tf.cast(
tf.constant(adjM[tuple([new_parts[i]])][:, new_parts[i]]),
'float32') for i in range(num_neurons)
]
stacked = [
tf.transpose(
tf.tensordot(stacked[i], adjMs_forNodes[i], axes=0),
self.tensordot_swap_channel_list)
for i in range(num_neurons)
]
#temp=[[tf.einsum(expression, stacked[i]) for expression in self.contractions_expressions] for i in range(num_neurons)]
qs = [
tf.stack([
tf.einsum(expression,
*([stacked[i]] + operator(len(new_parts[i]))))
for expression, operator in zip(
self.contractions_expressions,
self.contractions_add_operators)
],
axis=1) for i in range(num_neurons)
]
activations = [
tf.transpose(
self.nonlinearity(
tf.add(
tf.einsum(self.einsum_activation, self.W,
qs[neuron_ind]), self.bias)),
self.activation_swap_channels_list)
for neuron_ind in range(num_neurons)
]
return activations, adjM, new_parts
def keys(self):
return OrderedSet.union(self._default.keys(), self.__dict__.keys())
