init

.gitattributes

@@ -472,3 +472,42 @@ data/tweet_sentiment_test1_seed0/validation.jsonl filter=lfs diff=lfs merge=lfs
 data/tweet_sentiment_test2_seed0/train.jsonl filter=lfs diff=lfs merge=lfs -text
 data/tweet_sentiment_test3_seed2/train.jsonl filter=lfs diff=lfs merge=lfs -text
 data/tweet_sentiment/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed1/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed2/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed0/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed0/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed2/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed0/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed1/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed0/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed2/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed0/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed2/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed0/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed2/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed0/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed2/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed2/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed2/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed2/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed2/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed0/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed2/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed0/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed1/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed2/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed1/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed0/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed1/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed1/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed1/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test0_seed1/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed0/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test1_seed1/test.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed1/train.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test2_seed1/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed0/validation.jsonl filter=lfs diff=lfs merge=lfs -text
+data/tweet_sentiment_small_test3_seed1/train.jsonl filter=lfs diff=lfs merge=lfs -text

data/tweet_sentiment_small/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16b4c43d7af04f80c58c93a4b3da1984519ed0cd8ff78361885677cede78e003
+size 188450

data/tweet_sentiment_small/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7a4e19028ea5b3d4510d8b69d9af74a1821a74bfae16fd74d8b7856d79262a0e
+size 691715

data/tweet_sentiment_small/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9250261dcf45e5bca91a6757b29685036ce810708247cbe588a1a92963f4e85f
+size 187256

data/tweet_sentiment_small_test0_seed0/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09b915a486b09ab99730b2177ed285467e9db60f2c782768411d241d28f50a21
+size 47187

data/tweet_sentiment_small_test0_seed0/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb29b1640b8d040d84f499a7be85d8c0c1a2e0d5c219a96d7e59fee91e028283
+size 693690

data/tweet_sentiment_small_test0_seed0/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d455aa627a3f64075cfa6f92f46be1c70f2253b248bbbb862a0176cb4b75d3cc
+size 187256

data/tweet_sentiment_small_test0_seed1/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09b915a486b09ab99730b2177ed285467e9db60f2c782768411d241d28f50a21
+size 47187

data/tweet_sentiment_small_test0_seed1/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bad40d185e7e5091a38738d1c7ce9f5fb9c2bd3774a40891fd537af3e294294d
+size 693342

data/tweet_sentiment_small_test0_seed1/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28f36c1ff7cf7ebc22470f6441bc5687a5d38ee6fb3e51916db003b1ce21d607
+size 187256

data/tweet_sentiment_small_test0_seed2/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09b915a486b09ab99730b2177ed285467e9db60f2c782768411d241d28f50a21
+size 47187

data/tweet_sentiment_small_test0_seed2/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:827bd0ad88ef5b9513b2bf8e6c733d35adacec30d43bceae3dcf9d91b101b78c
+size 694096

data/tweet_sentiment_small_test0_seed2/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a73d92f1a7dec42909ae154175de9c21a5f6851d524b7691163ef4be3c8c3ef0
+size 187256

data/tweet_sentiment_small_test1_seed0/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bf7c917640a83060e28a6450999bca55da3fead2a522993c1c721380700ffc3
+size 47265

data/tweet_sentiment_small_test1_seed0/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:986866ae2e73151ae9b9205e0592c2a118c29a07c23e1551d83e0c77a3d1f266
+size 694027

data/tweet_sentiment_small_test1_seed0/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f94503184ee82f8f12d8c16fbcebdfcb915f74143643e26dd89a0a572b0eaa9
+size 187256

data/tweet_sentiment_small_test1_seed1/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bf7c917640a83060e28a6450999bca55da3fead2a522993c1c721380700ffc3
+size 47265

data/tweet_sentiment_small_test1_seed1/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a60a2f1ae7bdbf4f139477b17014f1133e2dd9bb2b3a60c30af5601245320d6
+size 694124

data/tweet_sentiment_small_test1_seed1/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d133c09396c40b114d755a07913d2d9562da3523cfd4fa61f9609867c0f38a3
+size 187256

data/tweet_sentiment_small_test1_seed2/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bf7c917640a83060e28a6450999bca55da3fead2a522993c1c721380700ffc3
+size 47265

data/tweet_sentiment_small_test1_seed2/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e9575bdadf4ae3de8eb354ea520b9ab5d35cf6eaaaa4a202ffda25b8c2788287
+size 693765

data/tweet_sentiment_small_test1_seed2/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf5fa46561883333a42ed602df67c22bb041bf342d3f227221ccd4c0b062fad0
+size 187256

data/tweet_sentiment_small_test2_seed0/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26bdf025b4e171101f303be4dcead3e75e5e7461124132abaab9ed7bdc6ac3c7
+size 47824

data/tweet_sentiment_small_test2_seed0/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c84b5be06a70297ade73615b4195373a16c34196847a1540b81f67bf365c3589
+size 692031

data/tweet_sentiment_small_test2_seed0/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ec01d63ba6f603aa1e40c7151fbb14c0557c27d37a76dea2fb712e0c155f83be
+size 187256

data/tweet_sentiment_small_test2_seed1/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26bdf025b4e171101f303be4dcead3e75e5e7461124132abaab9ed7bdc6ac3c7
+size 47824

data/tweet_sentiment_small_test2_seed1/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e2dea020a4572d6c4ca6c83861e0e0f7d847d5a9ca4205a8befa451679c610be
+size 692103

data/tweet_sentiment_small_test2_seed1/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2827b324e68b7c007aba908bff038b9416ae9ed250ca3bdde10d7943f9c00684
+size 187256

data/tweet_sentiment_small_test2_seed2/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26bdf025b4e171101f303be4dcead3e75e5e7461124132abaab9ed7bdc6ac3c7
+size 47824

data/tweet_sentiment_small_test2_seed2/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28429edc222382eb2b7407d2f09b52691724822baa0df07f7e252b6fa45cbbc1
+size 692384

data/tweet_sentiment_small_test2_seed2/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855657e67e61720f3c622c089a168674c8af516205f34e12c18576e73f2ddd1c
+size 187256

data/tweet_sentiment_small_test3_seed0/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:395be43c04fa28853568e1c0771e450e0aac1dfaf55b986ed77cde76662fb51e
+size 46171

data/tweet_sentiment_small_test3_seed0/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19392e2769e3fa011bc0e108dc578100d95ce315fd746355ef4c98ebd296c187
+size 694740

data/tweet_sentiment_small_test3_seed0/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b5c4bd6f073175119e1d158dcacfef4fce3119e5e881106134e196d452bebf5
+size 187256

data/tweet_sentiment_small_test3_seed1/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:395be43c04fa28853568e1c0771e450e0aac1dfaf55b986ed77cde76662fb51e
+size 46171

data/tweet_sentiment_small_test3_seed1/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df6c900a3411f5b7db03436bd862d5854f672bbe3a570142025c99bb6cbe23af
+size 694752

data/tweet_sentiment_small_test3_seed1/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:beab432ab7afb24e83df30b00f655f98ef457e6097a6ff82556a4a16fb47a285
+size 187256

data/tweet_sentiment_small_test3_seed2/test.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:395be43c04fa28853568e1c0771e450e0aac1dfaf55b986ed77cde76662fb51e
+size 46171

data/tweet_sentiment_small_test3_seed2/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c99e502259b679e06898b0b2711b5f319814284a9e2a68bbc2099897bcfefd3
+size 694703

data/tweet_sentiment_small_test3_seed2/validation.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cea3144fe17b4d74d2cd5628006adf248d218996c492f4567e57d94d7d4b5e3a
+size 187256

experiments/main.sh

@@ -1,13 +1,23 @@
-# topic, nerd[hw], ner, sentiment[hw]
+# MAIN EXPERIMENT
+MODEL="roberta-base"
 MODEL="vinai/bertweet-base"
-# topic, nerd[hw], ner[ukri], sentiment[hw]
 MODEL="cardiffnlp/twitter-roberta-base-2022-154m"
-# topic, nerd[ukri], ner[stone], sentiment[todo]
 MODEL="jhu-clsp/bernice"
-# topic, nerd[hw], ner[stone], sentiment[todo]
 MODEL="cardiffnlp/twitter-roberta-base-2021-124m"
-# topic, nerd[hw], ner[ukri], sentiment[todo]
-MODEL="roberta-base"
+
+# topic, ner, nerd, sentiment[stone]
+MODEL="roberta-large"
+# topic, ner, nerd[hk], sentiment[ukri]
+MODEL="vinai/bertweet-large"
+# topic, ner, nerd[hk], sentiment[ukri]
+MODEL="cardiffnlp/twitter-roberta-large-2022-154m"
+
+# ABLATION (TimeLMs)
+## Topic & NER
+MODEL="twitter-roberta-base-jun2020"
+MODEL="twitter-roberta-base-sep2021"
+## NERD
+MODEL="twitter-roberta-base-jun2021"
 
 # SENTIMENT
 python model_finetuning_sentiment.py -m "${MODEL}" -d "sentiment_temporal"
@@ -24,6 +34,8 @@ python model_finetuning_sentiment.py -m "${MODEL}" -d "sentiment_random1_seed2"
 python model_finetuning_sentiment.py -m "${MODEL}" -d "sentiment_random2_seed2"
 python model_finetuning_sentiment.py -m "${MODEL}" -d "sentiment_random3_seed2"
 
+python model_finetuning_sentiment.py -m "${MODEL}" -d "sentiment_random1_seed2"
+rm -rf ckpt
 # NERD
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_temporal"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random0_seed0"
@@ -34,7 +46,6 @@ python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random0_seed1"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random1_seed1"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random2_seed1"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random3_seed1"
-
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random0_seed2"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random1_seed2"
 python model_finetuning_nerd.py -m "${MODEL}" -d "nerd_random2_seed2"

process/tweet_sentiment.py

@@ -22,6 +22,7 @@ n_train = len(train)
 n_validation = len(validation)
 n_test = int(len(test)/4)
 
+
 def sampler(dataset_test, r_seed):
     seed(r_seed)
     shuffle(dataset_test)

process/tweet_sentiment_small.py

@@ -0,0 +1,71 @@
+# TODO
+import json
+import os
+from random import shuffle, seed
+
+
+with open("data/tweet_sentiment/test.jsonl") as f:
+    test = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/test_1.jsonl") as f:
+    test_1 = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/test_2.jsonl") as f:
+    test_2 = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/test_3.jsonl") as f:
+    test_3 = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/test_4.jsonl") as f:
+    test_4 = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/train.jsonl") as f:
+    train = [json.loads(i) for i in f if len(i)]
+with open("data/tweet_sentiment/validation.jsonl") as f:
+    validation = [json.loads(i) for i in f if len(i)]
+
+
+os.makedirs(f"data/tweet_sentiment_small", exist_ok=True)
+with open(f"data/tweet_sentiment_small/test.jsonl", "w") as f:
+    f.write("\n".join([json.dumps(i) for i in test]))
+with open(f"data/tweet_sentiment_small/validation.jsonl", "w") as f:
+    f.write("\n".join([json.dumps(i) for i in validation]))
+
+# down sample training set
+n_train_p = 2500
+n_train_n = 2500
+seed(123)
+shuffle(train)
+train_p = [i for i in train if i["gold_label_binary"] == 0][:n_train_p]
+train_n = [i for i in train if i["gold_label_binary"] == 1][:n_train_n]
+train = train_p + train_n
+shuffle(train)
+with open(f"data/tweet_sentiment_small/train.jsonl", "w") as f:
+    f.write("\n".join([json.dumps(i) for i in train]))
+
+
+n_train = len(train)
+n_validation = len(validation)
+n_test = int(len(test)/4)
+
+def sampler(dataset_test, r_seed):
+    seed(r_seed)
+    shuffle(dataset_test)
+    shuffle(train)
+    shuffle(validation)
+    test_tr = dataset_test[:int(n_train / 2)]
+    test_vl = dataset_test[int(n_train / 2): int(n_train / 2) + int(n_validation / 2)]
+    new_train = test_tr + train[:n_train - len(test_tr)]
+    new_validation = test_vl + validation[:n_validation - len(test_vl)]
+    return new_train, new_validation
+
+id2test = {n: t for n, t in enumerate([test_1, test_2, test_3, test_4])}
+for n, _test in enumerate([
+        test_4 + test_2 + test_3,
+        test_1 + test_4 + test_3,
+        test_1 + test_2 + test_4,
+        test_1 + test_2 + test_3]):
+    for s in range(3):
+        os.makedirs(f"data/tweet_sentiment_small_test{n}_seed{s}", exist_ok=True)
+        _train, _valid = sampler(_test, s)
+        with open(f"data/tweet_sentiment_small_test{n}_seed{s}/train.jsonl", "w") as f:
+            f.write("\n".join([json.dumps(i) for i in _train]))
+        with open(f"data/tweet_sentiment_small_test{n}_seed{s}/validation.jsonl", "w") as f:
+            f.write("\n".join([json.dumps(i) for i in _valid]))
+        with open(f"data/tweet_sentiment_small_test{n}_seed{s}/test.jsonl", "w") as f:
+            f.write("\n".join([json.dumps(i) for i in id2test[n]]))

statistics.py

@@ -25,3 +25,16 @@ for i in ["nerd_temporal", "ner_temporal", "topic_temporal", "sentiment_temporal
          })
 df = pd.DataFrame(stats)
 print(df)
+pretty_name = {
+    "nerd_temporal": "NERD",
+    "ner_temporal": "NER",
+    "topic_temporal": "Topic",
+    "sentiment_temporal": "Sentiment"
+}
+df.index = [pretty_name[i] for i in df.pop("data")]
+df = df[["split", "size", "date"]]
+pretty_name_split = {"train": "Train", "validation": "Valid", "test": "Test"}
+df["split"] = [pretty_name_split[i] for i in df["split"]]
+df.columns = [i.capitalize() for i in df.columns]
+df['Size'] = df['Size'].map('{:,}'.format)
+print(df.to_latex())

tweet_temporal_shift.py

@@ -124,6 +124,13 @@ class TweetTemporalShift(datasets.GeneratorBasedBuilder):
             features=["gold_label_binary", "text", "date"],
             data_url=f"{_ROOT_URL}/tweet_sentiment",
         ),
+        TweetTemporalShiftConfig(
+            name="sentiment_temporal_small",
+            description=_TWEET_SENTIMENT_DESCRIPTION,
+            citation=_TWEET_SENTIMENT_CITATION,
+            features=["gold_label_binary", "text", "date"],
+            data_url=f"{_ROOT_URL}/sentiment_temporal_small",
+        ),
     ]
     for s in range(3):
         for i in range(4):