init

experiments/huggingface_ops.py

@@ -3,7 +3,7 @@ from pprint import pprint
 
 api = HfApi()
 models = api.list_models(filter=ModelFilter(author='tweettemposhift'))
-models_filtered = [i.modelId for i in models if 'emoji-' in i.modelId]
+models_filtered = [i.modelId for i in models if 'hate_balance' in i.modelId]
 pprint(sorted(models_filtered))
 for i in models_filtered:
     api.delete_repo(i, repo_type="model")

experiments/main.sh

@@ -5,12 +5,12 @@ MODEL="jhu-clsp/bernice"
 MODEL="cardiffnlp/twitter-roberta-base-2019-90m"
 MODEL="cardiffnlp/twitter-roberta-base-dec2020"
 MODEL="cardiffnlp/twitter-roberta-base-2021-124m"
+MODEL="cardiffnlp/twitter-roberta-base-2022-154m"
+
 MODEL="roberta-large"
 MODEL="vinai/bertweet-large"
 MODEL="cardiffnlp/twitter-roberta-large-2022-154m"
 
-# todo
-MODEL="cardiffnlp/twitter-roberta-base-2022-154m"
 
 # EMOJI
 python model_finetuning_emoji.py -m "${MODEL}" -d "emoji_temporal"

process/tweet_hate_balance.py

@@ -38,6 +38,8 @@ def sampler(chunk_index, r_seed):
     df_train_tmp_n = df_train_negative.sample(label_train_dist[0] - label_train_tmp_dist[0], random_state=r_seed)
     df_train_tmp = pd.concat([df_train_tmp, df_train_tmp_p, df_train_tmp_n])
     assert dict(zip(*np.unique(df_train_tmp["gold_label_binary"], return_counts=True))) == label_train_dist
+    print(dict(zip(*np.unique(df_train_tmp["gold_label_binary"], return_counts=True))))
+    print(label_train_dist)
 
     label_val_tmp_dist = dict(zip(*np.unique(df_val_tmp["gold_label_binary"], return_counts=True)))
     assert label_val_tmp_dist[0] < label_val_dist[0] and label_val_tmp_dist[1] < label_val_dist[1]