def random_walk_error(sampling_args):
graph = sampling_args['graph']
samples = sampling.main(sampling_args)['sampling_set']
x_hat = sparse_label_propagation(graph, list(samples))
x = [graph.node[idx]['value'] for idx in range(graph.number_of_nodes())]
error = normalized_mean_squared_error(x, x_hat)
return error
def get_current_nmse(self):
graph = self.graph
sampling_set = self.sampling_set
x = [graph.node[i]['value'] for i in sorted(graph.nodes_iter())]
x_hat = sparse_label_propagation(graph, list(sampling_set))
return nmse(x, x_hat)
def _reward(self):
x_hat = sparse_label_propagation(self.graph, list(self.sampling_set))
x = [
self.graph.node[idx]['value']
for idx in range(self.graph.number_of_nodes())
]
error = nmse(x, x_hat)
reward = (self.slp_maximum_error - error) / self.slp_maximum_error
return reward
def get_current_nmse(self):
graph = self.graph
sampling_set = self.sampling_set
x = [
self.graph.node[idx]['value']
for idx in range(self.graph.number_of_nodes())
]
x_hat = sparse_label_propagation(graph, list(sampling_set))
return nmse(x, x_hat)
def slp_minimum_error(graph, sampling_set_size):
"""Finds the minimum error for given graph and sampling set size.
Very slow for large graphs due to exhaustive search.
"""
sampling_sets = combinations(graph.nodes(), sampling_set_size)
x = [graph.node[idx]['value'] for idx in range(graph.number_of_nodes())]
errors = []
for sampling_set in sampling_sets:
x_hat = sparse_label_propagation(graph, sampling_set)
error = normalized_mean_squared_error(x, x_hat)
errors.append(error)
return min(errors)
def _reward(self):
x_hat = sparse_label_propagation(self.graph, list(self.sampling_set))
x = [
self.graph.node[idx]['value']
for idx in range(self.graph.number_of_nodes())
]
error = nmse(x, x_hat)
tv = total_variation(self.graph.edges(), x)
self.error = error
# this + all nodes in action seems best
reward = 1.0 - error
return reward