Fix bug and checksum in irc_disentangle dataset (#4377)

* Fix filepath segment and refactor

* Update checksum in metadata JSON

* Fix style

* Update dummy data

Commit from https://github.com/huggingface/datasets/commit/f2ef5e81822f976afe9baf21d0e6697793b2c960

dataset_infos.json

@@ -1 +1 @@
-{"ubuntu": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large, \n title = \"A Large-Scale Corpus for Conversation Disentanglement\", \n author = \"Kummerfeld, Jonathan K.  and \n Gouravajhala, Sai R.  and \n Peper, Joseph J.  and \n Athreya, Vignesh  and \n Gunasekara, Chulaka  and \n Ganhotra, Jatin  and \n Patel, Siva Sankalp  and \n Polymenakos, Lazaros C  and \n Lasecki, Walter\", \n booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\", \n month = jul, \n year = \"2019\", \n address = \"Florence, Italy\", \n publisher = \"Association for Computational Linguistics\", \n url = \"https://aclanthology.org/P19-1374\", \n doi = \"10.18653/v1/P19-1374\", \n pages = \"3846--3856\", \n arxiv = \"https://arxiv.org/abs/1810.11118\", \n software = \"https://jkk.name/irc-disentanglement\", \n data = \"https://jkk.name/irc-disentanglement\", \n abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\", \n}", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "date": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "irc_disentangle", "config_name": "ubuntu", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 55970472, "num_examples": 220616, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 3916881, "num_examples": 15010, "dataset_name": "irc_disentangle"}, "validation": {"name": "validation", "num_bytes": 3079360, "num_examples": 12510, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470349, "checksum": "9e2c98a15191d729c0dbe10f309d836bd1ab32c0d03ffb2e0e4205287405fc4d"}}, "download_size": 118470349, "post_processing_size": null, "dataset_size": 62966713, "size_in_bytes": 181437062}, "channel_two": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large, \n title = \"A Large-Scale Corpus for Conversation Disentanglement\", \n author = \"Kummerfeld, Jonathan K.  and \n Gouravajhala, Sai R.  and \n Peper, Joseph J.  and \n Athreya, Vignesh  and \n Gunasekara, Chulaka  and \n Ganhotra, Jatin  and \n Patel, Siva Sankalp  and \n Polymenakos, Lazaros C  and \n Lasecki, Walter\", \n booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\", \n month = jul, \n year = \"2019\", \n address = \"Florence, Italy\", \n publisher = \"Association for Computational Linguistics\", \n url = \"https://aclanthology.org/P19-1374\", \n doi = \"10.18653/v1/P19-1374\", \n pages = \"3846--3856\", \n arxiv = \"https://arxiv.org/abs/1810.11118\", \n software = \"https://jkk.name/irc-disentanglement\", \n data = \"https://jkk.name/irc-disentanglement\", \n abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\", \n}", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "builder_name": "irc_disentangle", "config_name": "channel_two", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"dev": {"name": "dev", "num_bytes": 197189, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot": {"name": "pilot", "num_bytes": 92514, "num_examples": 501, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 186494, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot_dev": {"name": "pilot_dev", "num_bytes": 289695, "num_examples": 1501, "dataset_name": "irc_disentangle"}, "all_": {"name": "all_", "num_bytes": 495666, "num_examples": 2602, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470349, "checksum": "9e2c98a15191d729c0dbe10f309d836bd1ab32c0d03ffb2e0e4205287405fc4d"}}, "download_size": 118470349, "post_processing_size": null, "dataset_size": 1261558, "size_in_bytes": 119731907}}
+{"ubuntu": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large,\n    title = \"A Large-Scale Corpus for Conversation Disentanglement\",\n    author = \"Kummerfeld, Jonathan K.  and\n      Gouravajhala, Sai R.  and\n      Peper, Joseph J.  and\n      Athreya, Vignesh  and\n      Gunasekara, Chulaka  and\n      Ganhotra, Jatin  and\n      Patel, Siva Sankalp  and\n      Polymenakos, Lazaros C  and\n      Lasecki, Walter\",\n    booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\",\n    month = jul,\n    year = \"2019\",\n    address = \"Florence, Italy\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://aclanthology.org/P19-1374\",\n    doi = \"10.18653/v1/P19-1374\",\n    pages = \"3846--3856\",\n    arxiv = \"https://arxiv.org/abs/1810.11118\",\n    software = \"https://jkk.name/irc-disentanglement\",\n    data = \"https://jkk.name/irc-disentanglement\",\n    abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\",\n}\n", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "date": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "irc_disentangle", "config_name": "ubuntu", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 56012854, "num_examples": 220616, "dataset_name": "irc_disentangle"}, "validation": {"name": "validation", "num_bytes": 3081479, "num_examples": 12510, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 3919900, "num_examples": 15010, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470210, "checksum": "e5232a65f5e97805a366a19b4c0b127dfcf91981a8681b33855bfb6c72706c2f"}}, "download_size": 118470210, "post_processing_size": null, "dataset_size": 63014233, "size_in_bytes": 181484443}, "channel_two": {"description": "Disentangling conversations mixed together in a single stream of messages is\na difficult task, made harder by the lack of large manually annotated\ndatasets. This new dataset of 77,563 messages manually annotated with\nreply-structure graphs that both disentangle conversations and define\ninternal conversation structure. The dataset is 16 times larger than all\npreviously released datasets combined, the first to include adjudication of\nannotation disagreements, and the first to include context.\n", "citation": "@inproceedings{kummerfeld-etal-2019-large,\n    title = \"A Large-Scale Corpus for Conversation Disentanglement\",\n    author = \"Kummerfeld, Jonathan K.  and\n      Gouravajhala, Sai R.  and\n      Peper, Joseph J.  and\n      Athreya, Vignesh  and\n      Gunasekara, Chulaka  and\n      Ganhotra, Jatin  and\n      Patel, Siva Sankalp  and\n      Polymenakos, Lazaros C  and\n      Lasecki, Walter\",\n    booktitle = \"Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics\",\n    month = jul,\n    year = \"2019\",\n    address = \"Florence, Italy\",\n    publisher = \"Association for Computational Linguistics\",\n    url = \"https://aclanthology.org/P19-1374\",\n    doi = \"10.18653/v1/P19-1374\",\n    pages = \"3846--3856\",\n    arxiv = \"https://arxiv.org/abs/1810.11118\",\n    software = \"https://jkk.name/irc-disentanglement\",\n    data = \"https://jkk.name/irc-disentanglement\",\n    abstract = \"Disentangling conversations mixed together in a single stream of messages is a difficult task, made harder by the lack of large manually annotated datasets. We created a new dataset of 77,563 messages manually annotated with reply-structure graphs that both disentangle conversations and define internal conversation structure. Our data is 16 times larger than all previously released datasets combined, the first to include adjudication of annotation disagreements, and the first to include context. We use our data to re-examine prior work, in particular, finding that 89% of conversations in a widely used dialogue corpus are either missing messages or contain extra messages. Our manually-annotated data presents an opportunity to develop robust data-driven methods for conversation disentanglement, which will help advance dialogue research.\",\n}\n", "homepage": "https://jkk.name/irc-disentanglement/", "license": "Creative Commons Attribution 4.0 International Public License", "features": {"id": {"dtype": "int32", "id": null, "_type": "Value"}, "raw": {"dtype": "string", "id": null, "_type": "Value"}, "ascii": {"dtype": "string", "id": null, "_type": "Value"}, "tokenized": {"dtype": "string", "id": null, "_type": "Value"}, "connections": {"feature": {"dtype": "int32", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "irc_disentangle", "config_name": "channel_two", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"dev": {"name": "dev", "num_bytes": 197505, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot": {"name": "pilot", "num_bytes": 92663, "num_examples": 501, "dataset_name": "irc_disentangle"}, "test": {"name": "test", "num_bytes": 186823, "num_examples": 1001, "dataset_name": "irc_disentangle"}, "pilot_dev": {"name": "pilot_dev", "num_bytes": 290175, "num_examples": 1501, "dataset_name": "irc_disentangle"}, "all_": {"name": "all_", "num_bytes": 496524, "num_examples": 2602, "dataset_name": "irc_disentangle"}}, "download_checksums": {"https://github.com/jkkummerfeld/irc-disentanglement/tarball/master": {"num_bytes": 118470210, "checksum": "e5232a65f5e97805a366a19b4c0b127dfcf91981a8681b33855bfb6c72706c2f"}}, "download_size": 118470210, "post_processing_size": null, "dataset_size": 1263690, "size_in_bytes": 119733900}}

dummy/channel_two/1.0.0/dummy_data.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a1b5c2585f5c49e8c4266399d52f674dd46baa2adf9abe11c8580b7bd4d469b
-size 14669
+oid sha256:f12c3e7bf0122ac366937f68f7e490839d7e95ab303e09bfaacd61b7e5ffb812
+size 9878

dummy/ubuntu/1.0.0/dummy_data.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:81944d1b2e274034c077343f94b9cc292bea0e3664050c6fddabbc6fbf5abab3
-size 16543
+oid sha256:42375070ad386008d4500192af4d3a0e0fc36d9e951a38884f9b185b954df2da
+size 9194

irc_disentangle.py

@@ -119,40 +119,22 @@ class IRCDisentangle(datasets.GeneratorBasedBuilder):
 
     def _split_generators(self, dl_manager):
         """Returns SplitGenerators."""
-        my_urls = _URL
-        dl_dir = dl_manager.download_and_extract(my_urls)
-
-        files = dict()
+        dl_dir = dl_manager.download_and_extract(_URL)
+        filepath = os.path.join(dl_dir, "jkkummerfeld-irc-disentanglement-35f0a40", "data")
+        split_names = {datasets.Split.TRAIN: "train", datasets.Split.VALIDATION: "dev", datasets.Split.TEST: "test"}
         if self.config.name == "ubuntu":
-            for split in ["train", "dev", "test"]:
-                files[split] = os.path.join(dl_dir, "jkkummerfeld-irc-disentanglement-fd379e9", "data", split)
-
             return [
                 datasets.SplitGenerator(
-                    name=datasets.Split.TRAIN,
-                    gen_kwargs={
-                        "filepath": files["train"],
-                        "split": "train",
-                    },
-                ),
-                datasets.SplitGenerator(
-                    name=datasets.Split.TEST,
-                    gen_kwargs={
-                        "filepath": files["test"],
-                        "split": "test",
-                    },
-                ),
-                datasets.SplitGenerator(
-                    name=datasets.Split.VALIDATION,
+                    name=split,
                     gen_kwargs={
-                        "filepath": files["dev"],
-                        "split": "dev",
+                        "filepath": os.path.join(filepath, split_name),
+                        "split": split_name,
                     },
-                ),
+                )
+                for split, split_name in split_names.items()
             ]
-
         elif self.config.name == "channel_two":
-            filepath = os.path.join(dl_dir, "jkkummerfeld-irc-disentanglement-fd379e9", "data", "channel-two")
+            filepath = os.path.join(filepath, "channel-two")
             return [
                 datasets.SplitGenerator(
                     name="dev",