def infer(self, raw_graph_data, **kwargs):
graphs = [self.process_raw_graph(g) for g in raw_graph_data]
batch_iterator = utils.ThreadedIterator(self.make_minibatch_iterator(
graphs, is_training=False),
max_queue_size=50)
preds = []
for step, batch_data in enumerate(batch_iterator):
batch_data[self.placeholders['out_layer_dropout_keep_prob']] = 1.0
fetch_list = [
self.ops['probabilities'],
self.placeholders['elem_graph_nodes_list'],
self.placeholders['elements_true']
]
result = self.sess.run(fetch_list, feed_dict=batch_data)
groupings = collections.defaultdict(list)
for (discard_prob, keep_prob), graph_id, candidate in zip(*result):
groupings[graph_id].append(
(discard_prob, keep_prob, len(groupings[graph_id])))
preds += list(groupings.values())
return preds
def infer(self, raw_graph_data, **kwargs):
graphs = [self.process_raw_graph(g) for g in raw_graph_data]
batch_iterator = utils.ThreadedIterator(self.make_minibatch_iterator(graphs, is_training=False),
max_queue_size=50)
preds = []
for step, batch_data in enumerate(batch_iterator):
batch_data[self.placeholders['out_layer_dropout_keep_prob']] = 1.0
fetch_list = [self.ops['preds'], self.ops['probs']]
result = self.sess.run(fetch_list, feed_dict=batch_data)
for p, v in zip(result[1], result[0]):
preds.append(list(zip(p, v)))
return preds
def infer(self, raw_graph_data, **kwargs):
graphs = [self.process_raw_graph(g) for g in raw_graph_data]
graphs = [g for g in graphs if g is not None]
batch_iterator = utils.ThreadedIterator(self.make_minibatch_iterator(
graphs, is_training=False),
max_queue_size=50)
preds = []
max_length = self.params['max_length']
for step, batch_data in enumerate(batch_iterator):
batch_data[self.placeholders['out_layer_dropout_keep_prob']] = 1.0
fetch_list = [
self.ops['softmax_values'],
self.placeholders['graph_ids_timestep'],
self.placeholders['elements_true']
]
result = self.sess.run(fetch_list, feed_dict=batch_data)
softmax_values, graph_ids, elements = result
elements = np.concatenate([elements] * max_length, axis=0)
groupings = collections.defaultdict(list)
for probs, graph_id, element in zip(softmax_values, graph_ids,
elements):
groupings[graph_id].append((probs, element))
for graph_id, flat_preds in groupings.items():
num_elems = len(flat_preds) // max_length
graph_preds = []
for i in range(0, len(flat_preds), num_elems):
graph_preds.append(
sorted(flat_preds[i:i + num_elems],
key=lambda x: -x[0]))
preds.append(graph_preds)
return preds
def analyze(self, test_data):
with self.graph.as_default():
batch_iterator = utils.ThreadedIterator(
self.make_minibatch_iterator(test_data, False),
max_queue_size=5)
fetch_list = self.get_fetch_list()
processed_graphs = 0
for step, batch_data in enumerate(batch_iterator):
batch_data[
self.placeholders['out_layer_dropout_keep_prob']] = 1.0
num_graphs = batch_data[
self.placeholders['num_graphs_in_batch']]
processed_graphs += num_graphs
print("Running Analysis, batch {} (has {} graphs).".format(
step, num_graphs),
end='\r')
result = self.sess.run(fetch_list, feed_dict=batch_data)
self.analyze_result(num_graphs, result)
print("\n-----\n")
return self.finish_analysis()