def create_frame(intent_label, slot_names_str, utterance):
frame = Node(
label=intent_label,
span=Span(0, byte_length(utterance)),
children={
Node(label=slot.label, span=Span(slot.start, slot.end))
for slot in parse_slot_string(slot_names_str)
},
)
return frame
def aggregate_targets(self, batch_targets, batch_context):
intent_targets = batch_targets[0]
self.all_targets.extend([
create_frame(
text,
self.doc_label_names[intent_target],
raw_slot_label,
byte_length(text),
) for text, intent_target, raw_slot_label, seq_len in zip(
batch_context[self.text_column_name],
intent_targets,
batch_context[DatasetFieldName.RAW_WORD_LABEL],
batch_context[DatasetFieldName.SEQ_LENS],
)
])
def test_create_node(self):
TEXT_EXAMPLES = [
("exit", "device/close_app", "", "[device/close_app exit ]"),
(
"call mom",
"IN:CREATE_CALL",
"5:8:SL:CONTACT",
"[IN:CREATE_CALL call [SL:CONTACT mom ] ]",
),
(
"An Yu",
"meta/provideSlotValue",
"0:5:SL:CONTACT",
"[meta/provideSlotValue [SL:CONTACT An Yu ] ]",
),
(
"Set a reminder to pick up Sean at 3:15 pm today.",
"IN:CREATE_REMINDER",
"18:30:SL:TODO,34:47:SL:DATE_TIME",
"[IN:CREATE_REMINDER Set a reminder to [SL:TODO pick up Sean ] "
"at [SL:DATE_TIME 3:15 pm today ]. ]",
),
(
"Set a reminder to pick up Sean at 3:15 pm today.",
"IN:CREATE_REMINDER",
"34:47:SL:DATE_TIME,18:30:SL:TODO",
"[IN:CREATE_REMINDER Set a reminder to [SL:TODO pick up Sean ] "
"at [SL:DATE_TIME 3:15 pm today ]. ]",
),
('["Fine"]', "cu:other", "", r'[cu:other \["Fine"\] ]'),
( # Example in byte offset.
"establece el escándalo a 7",
"IN:SET_VOLUME",
"26:27:SL:PRECISE_AMOUNT",
"[IN:SET_VOLUME establece el escándalo a [SL:PRECISE_AMOUNT 7 ] ]",
),
]
for (
utterance,
intent_label,
slot_names_str,
expected_annotation_str,
) in TEXT_EXAMPLES:
frame = create_frame(
utterance, intent_label, slot_names_str, byte_length(utterance)
)
self.assertEqual(frame_to_str(frame), expected_annotation_str)
def aggregate_preds(self, batch_preds, batch_context):
intent_preds, word_preds = batch_preds
self.all_preds.extend([
create_frame(
text,
self.doc_label_names[intent_pred],
merge_token_labels_to_slot(
token_range[0:seq_len],
[self.word_label_names[p] for p in word_pred[0:seq_len]],
self.use_bio_labels,
),
byte_length(text),
) for text, intent_pred, word_pred, seq_len, token_range in zip(
batch_context[self.text_column_name],
intent_preds,
word_preds,
batch_context[DatasetFieldName.SEQ_LENS],
batch_context[DatasetFieldName.TOKEN_RANGE],
)
])