Update etpc.py

etpc.py

@@ -19,6 +19,7 @@ import os
 from typing import Any, Dict, Generator, List, Optional, Tuple, Union
 
 import datasets
+import numpy as np
 from datasets.tasks import TextClassification
 from lxml import etree
 
@@ -86,8 +87,12 @@ class ETPC(datasets.GeneratorBasedBuilder):
                 "sentence2_segment_location": datasets.Sequence(
                     datasets.Value("int32")
                 ),
-                "sentence1_segment_text": datasets.Value("string"),
-                "sentence2_segment_text": datasets.Value("string"),
+                "sentence1_segment_text": datasets.Sequence(
+                    datasets.Value("string")
+                ),
+                "sentence2_segment_text": datasets.Sequence(
+                    datasets.Value("string")
+                ),
             }
         )
 
@@ -148,33 +153,43 @@ class ETPC(datasets.GeneratorBasedBuilder):
             sentence2_segment_text = root_paraphrase_types.xpath(
                 f".//pair_id[text()='{current_pair_id}']/parent::relation/s2_text/text()"
             )
-            sentence1_tokenized = row.find(".//sent1_tokenized").text.split()
-            sentence2_tokenized = row.find(".//sent2_tokenized").text.split()
 
-            sentence1_segment_location_full = [0] * len(sentence1_tokenized)
-            sentence2_segment_location_full = [0] * len(sentence2_tokenized)
+            sentence1_tokenized = row.find(".//sent1_tokenized").text.split(
+                " "
+            )
+            sentence2_tokenized = row.find(".//sent2_tokenized").text.split(
+                " "
+            )
+
+            sentence1_segment_location_full = np.zeros(
+                len(sentence1_tokenized)
+            )
+            sentence2_segment_location_full = np.zeros(
+                len(sentence2_tokenized)
+            )
 
             for (
-                sentence1_segment_location_example,
-                sentence2_segment_location_example,
+                sentence1_segment_locations,
+                sentence2_segment_locations,
                 paraphrase_type_id,
             ) in zip(
                 sentence1_segment_location,
                 sentence2_segment_location,
                 paraphrase_type_ids,
             ):
-                for (
-                    segment_location
-                ) in sentence1_segment_location_example.split(","):
-                    sentence1_segment_location_full[
-                        int(segment_location)
-                    ] = int(paraphrase_type_id)
-                for (
-                    segment_location
-                ) in sentence2_segment_location_example.split(","):
-                    sentence2_segment_location_full[
-                        int(segment_location)
-                    ] = int(paraphrase_type_id)
+                segment_locations_1 = [
+                    int(i) for i in sentence1_segment_locations.split(",")
+                ]
+                sentence1_segment_location_full[segment_locations_1] = [
+                    paraphrase_type_id
+                ] * len(segment_locations_1)
+
+                segment_locations_2 = [
+                    int(i) for i in sentence2_segment_locations.split(",")
+                ]
+                sentence2_segment_location_full[segment_locations_2] = [
+                    paraphrase_type_id
+                ] * len(segment_locations_2)
 
             yield idx, {
                 "idx": row.find(".//pair_id").text + "_" + str(idx),
@@ -192,5 +207,4 @@ class ETPC(datasets.GeneratorBasedBuilder):
                 "sentence1_segment_text": sentence1_segment_text,
                 "sentence2_segment_text": sentence2_segment_text,
             }
-
             idx += 1