MER-Baseline中修改分离出的文本单模态情感识别模块
.
├── config_path.py # 原始数据处理所用的路径
├── config.py # # 处理后数据集路径
├── extract_text_dataset.py # 原始数据集ASR，Text单模态数据集构造
├── extract_text_embedding_LZ.py # 从Text数据集提取embedding并存储
├── train.py # 分类网络和模型构建及训练
├── README.md
├── dataset # [Text单模态数据集]
├── result # [Text单模态TER结果]
├── features # gitignored: [embedding特征]
├── saved-unimodal # gitignored: [训练超参数等]
└── tools # gitignored: [预训练模型及工具库]
run：
# 提取embedding
python extract_text_embedding_LZ.py --dataset='MER2023' --feature_level='UTTERANCE' --model_name='chinese-macbert-large' --gpu=0# 文本单模态predict网络
python -u train.py --dataset='MER2023' --test_sets='test3' --audio_feature='chinese-macbert-large-4-UTT' --text_feature='chinese-macbert-large-4-UTT' --video_feature='chinese-macbert-large-4-UTT' --lr=1e-3 --gpu=0paper_result：
result：
# train_args_path为该次训练超参数保存path
fscore: 0.4377, valmse: 2.4225, metric: -0.1678,
train_args_path: ./saved-unimodal/model/cv_features:chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT_f1:0.4378_valmse:2.4225_metric:-0.1679_1683700986.6242754.npz
fscore: 0.4159, valmse: 2.4419, metric: -0.1945,
train_args_path: ./saved-unimodal/model/cv_features:chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT_f1:0.4159_valmse:2.4419_metric:-0.1945_1683701321.3115225.npz
fscore: 0.4261, valmse: 2.3998, metric: -0.1739,
train_args_path: ./saved-unimodal/model/cv_features:chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT+chinese-macbert-large-4-UTT_f1:0.4261_valmse:2.3998_metric:-0.1739_1683701999.5011146.npzrun：
# 提取embedding
python extract_text_embedding_LZ.py --dataset='MER2023' --feature_level='UTTERANCE' --model_name='chinese-roberta-wwm-ext-large' --gpu=0# 文本单模态predict网络
python -u train.py --dataset='MER2023' --test_sets='test3' --audio_feature='chinese-roberta-wwm-ext-large-4-UTT' --text_feature='chinese-roberta-wwm-ext-large-4-UTT' --video_feature='chinese-roberta-wwm-ext-large-4-UTT' --lr=1e-3 --gpu=0paper_result：
result：
# train_args_path为该次训练超参数保存path
fscore: 0.4245, valmse: 2.3955, metric: -0.1744,
train_args_path: ./saved-unimodal/model/cv_features:chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT_f1:0.4245_valmse:2.3955_metric:-0.1744_1683703252.0435195.npz
fscore: 0.4247, valmse: 2.3798, metric: -0.1703,
train_args_path: ./saved-unimodal/model/cv_features:chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT_f1:0.4247_valmse:2.3798_metric:-0.1703_1683703431.8806338.npz
fscore: 0.4237, valmse: 2.3435, metric: -0.1622,
train_args_path: ./saved-unimodal/model/cv_features:chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT+chinese-roberta-wwm-ext-large-4-UTT_f1:0.4237_valmse:2.3435_metric:-0.1622_1683703848.208033.npz