release-note.json

{
  "1.6.7(8)": {
    "1": "Fix a potential problem while do batch inference after training based on cached dataset"
  },
  "1.6.4": {
    "1": "Fix a potential problem in generate APC inference set",
    "2": "Register a Yelp dataset in PyABSA provided by WeLi9811: https://github.com/WeiLi9811"
  },
  "1.6.3": {
    "1": "Fix a potential problem in the sentiment classifier while loading tokenizer",
    "2": "Fix a problem in preprocessing in APC inference"
  },
  "1.6": {
    "0": "This is a stable version which eliminates almost all unknown problems before",
    "1": "Fix a problem in saving fine-tuned pretrained model",
    "2": "Fix an inference problem in LCA-BERT, SSW-T, SSW-S models",
    "3": "Fix a problem in updating ABSADatasets version",
    "4": "Modify the output format for BERTBaseline models and GloVe based models to adapt apc_trainer architecture",
    "5": "Fix the data preprocessing code for BERTBaseline models and GloVe based models",
    "6": "Rename the dependency matrix cache folder, that is for remove conflict between dependency matrix folder and integrated datasets",
    "7": "Add alert while loading fine-tuned models",
    "8": "Fix the inference for DLCF-DCA and DLCFS-DCA models",
    "9": "Fix the embedding function in IAN-Bert model",
    "10": "Fix the hop arg missing problem in Memnet-Bert, Ram-Bert, Memnet-GloVe, Ram-GloVe models",
    "11": "Fix a printing problem in baseline APC model inference result",
    "12": "Fix a parallel problem in BERT-BASE-ATEPC model",
    "13": "Stabilize the text_classifier",
    "14": "Fix test_loader init problem and dataset cache problem in ClassificationTrainer"
  },
  "1.5.4": {
    "1": "Fix a a bug while using custom dataset"
  },
  "1.5.3": {
    "1": "Minor fixes and modifications"
  },
  "1.5": {
    "1": "Release after full test, no known error yet",
    "2": "Fix BERT-SPC Modeling",
    "3": "Remove release-note check for efficient",
    "4": "Fix dataset cache function",
    "5": "Remove DistributedDataParallel for stability",
    "6": "Remove older checkpoints",
    "7": "Optimize early stop strategy, now patience means patience for epochs",
    "8": "Fix a problem may fail APC checkpoint loading",
    "9": "Fix a data loading problem for ATEPC",
    "10": "Add cache dataset option for all models",
    "99": "Other modifications"
  },
  "1.3.15": {
    "1": "Improve stability while using other pretrained models for ATEPC",
    "2": "This is a general update of default config for ATEPC"
  },
  "1.3.13": {
    "1": "Fix a checkpoint loading problem of APC (Some checkpoint at Google Drive may be unavailable now, we will update soon)"
  },
  "1.3.12": {
    "1": "Refactor to support customize IOB label for ATEPC (The integrated function to covert APC dataset to ATEPC dataset remains only support ASP IOB now, please customize your dataset's IOB label using your own script)",
    "2": "Update default pretrained model for ATEPC",
    "3": "Minor changes"
  },
  "1.3.11": {
    "1": "Divide LSA into FAST-LSA and LSA models"
  },
  "1.3.9": {
    "1": "Fix low performance of APC using roberta-base"
  },
  "1.3.8": {
    "1": "Fix a fatal problem in ATEPC example preprocessing (influenced versions: V1.X - V1.3.5), which triggers tremendous ASPECT TOO LONG WARNING. This error severely damaged the ATEPC performance. The ATEPC checkpoints on GoogleDrive were also influenced and may be updated in teh future. ",
    "2": "Add deep_ensemble option, use config.deep_ensemble=True to activate",
    "3": "Add early stop option, default patience=5",
    "4": "Refactor utils to print sorted args",
    "5": "Fix a problem while using checkpoint_save_mode=3 to save finetuned BERT",
    "6": "Refactor to retry training only while catching ConnectionError",
    "7": "Fix an ensemble problem in APC",
    "8": "Add full support distributed training",
    "9": "Add distributed training option i.e., DataParallel or DistributedDataParallel",
    "10": "Fix some potential problem in using other pretrained models in ATEPC to infer (caused by hard code [CLS], [SEP]), support roberta now",
    "11": "This is an public test version, could be removed later. Please roll back if you find any problem. I am sorry for my mistake, but I dont have enough time to maintain this project."
  },
  "1.3.5": {
    "1": "Update default pretrained_bert (bert-base-uncased -> roberta-base)",
    "2": "Add cache dataset option for APC task, use config.cache_dataset=True to activate"
  },
  "1.3.4": {
    "1": "Replace remaining BertModel.from_pretrained() and BertTokenizer.from_pretrained() with AutoModel.from_pretrained() and AutoTokenizer.from_pretrained()",
    "2": "Fix some ensemble problems"
  },
  "1.3.1": {
    "1": "Add multi-cuda support for APC model and part of ATEPC models",
    "2": "Add ensemble support for APC models",
    "3": "Fix support of legacy APC models in loading & inference using shared checkpoint "
  },
  "1.2.13": {
    "1": "Minor update in dataset searching"
  },
  "1.2.12": {
    "1": "Add set/get functions for configs"
  },
  "1.2.10": {
    "1": "Add an rule on APC dataset lines checking",
    "2": "Add SpaCy model config in classic APC models",
    "3": "Not fully tested for all situations"
  },
  "1.2.9": {
    "1": "Add an rule on APC dataset lines checking",
    "2": "You can use multiple types of label in your dataset, e.g., string, number"
  },
  "1.2.8": {
    "1": "Fix the convert_apc_set_to_atepc_set function",
    "2": "Fix the error to load a inference model from training, i.e., use trainer.load_trained_model() to load the inference model",
    "3": "Fix a bug of batch size setting in atepc inference",
    "4": "Fix a bug of auto label-mapping"
  },
  "1.2.7": {
    "1": "Deprecated"
  },
  "1.2.4": {
    "1": "Refactor checkpoint map processing format",
    "2": "Refactor APC inference to merge results with same text",
    "3": "Improve stability"
  },
  "1.2.3": {
    "1": "Enhance ATEPC dataset converting",
    "2": "Fix some problems in some particular situations",
    "3": "Improve stability"
  },
  "1.2.2": {
    "1": "Full support of flexible datasets, update ABSADatasets to version 2021.10.02",
    "2": "Support batch_size setting in ATEPC, APC, TC inference",
    "3": "Fix the inference of DLCF_DCA model",
    "4": "This version is for replacing 1.2.0(1)"
  },
  "1.2.0": {
    "1": "Enhance to support more flexible labels, now you can define string-based label instead of integer labels",
    "2": "Remove set_sentiment_map() support due to above modification",
    "3": "Fix a problem may cause problem while building graph for combined datasets",
    "4": "Fix a printing problem in ATEPC",
    "5": "Fix a bug in inference set loading",
    "6": "Redefine the Chinese datasets",
    "7": "This version involves considerable modification and may contain potential bug"
  },
  "1.1.24": {
    "1": "Add the parameters statistics function",
    "2": "Optimize the DLCF_DCA model"
  },
  "1.1.23": {
    "1": "Improve atepc aspect_extractor result, ensure final output is same length and order as original input examples",
    "2": "Fix a problem may merge all aspects of different example into 1 line if duplicate example is fed",
    "3": "Fix a problem may cause error in text classification",
    "4": "Fix a dataset loading problem"
  },
  "1.1.22": {
    "1": "Improve dataset search to be more flexible",
    "2": "Refactor label-mapping trigger. This feature is developed based on the mooc dataset: https://github.com/jmc-123/ABSADatasets/tree/master/datasets/apc_datasets/Chinese/mooc",
    "3": "Fix the batch inference of text classification",
    "4": "Fix the text classification dataset downloading problem",
    "5": "Fix a problem may cause failure of ATEPC inference",
    "6": "Add the dependency declaration of typing_extensions"
  },
  "1.1.20": {
    "1": "Add automatic ABSADatasets update check"
  },
  "1.1.19": {
    "1": "Fix training without testset in APC",
    "2": "Add SpaCy model setting option, e.g., config.spacy_model = 'zh_core_web_sm'"
  },
  "1.1.18": {
    "1": "Reformat and fix a bug of ATEPC output"
  },
  "1.1.17": {
    "1": "Add a new Chinese shampoo dataset, thanks to brightgems@github",
    "2": "Upgrade ABSADatasets to version: 2021.09.21",
    "3": "Fix the inference of DLCF_DCA",
    "4": "Fix the training and inferring LCA-Net model",
    "5": "Improve the config check function"
  },
  "1.1.16": {
    "1": "Enable flexible dataset format for ATEPC dataset, see https://github.com/yangheng95/PyABSA/issues/78",
    "2": "Fix a bug may cause checkpoint loading problem on no-cuda device",
    "3": "Add package version validation"
  },
  "1.1.14": {
    "1": "Fix the dataset processing functions"
  },
  "1.1.13": {
    "1": "Refactor ATEPC inference code",
    "2": "Add batch inference for APC and ATEPC, temporarily using freeze batch size",
    "3": "Define the English dataset"
  },
  "1.1.12": {
    "1": "Enable downloading shared checkpoint from a google drive url, this is for downloading checkpoint not registered in PyABSA",
    "2": "Refine LCF vec memory occupation",
    "3": "Add LCF-BERT2 and LCFS-BERT2 demo models",
    "4": "Fix a bug report (https://github.com/yangheng95/PyABSA/issues/73)"
  },
  "1.1.9": {
    "1": "Fix a problem in BERT-ATEPC model"
  },
  "1.1.8": {
    "1": "Fix a problem may cause checkpoint saving failure"
  },
  "1.1.7": {
    "1": "Fix the inference of ATEPC using internal datasets, if you are using 1.1.5 or 1.1.6, please update to this version",
    "2": "Improve stability and test all examples"
  },
  "1.1.6": {
    "1": "Deprecated"
  },
  "1.1.3": {
    "1": "Fix the feature to resume/retrain from a checkpoint"
  },
  "1.1.2": {
    "1": "Fix https://github.com/yangheng95/PyABSA/issues/59#issuecomment-902531502"
  },
  "1.1": {
    "1": "Fix some problems"
  },
  "1.0.7(.1.2.3)": {
    "1": "Fix all examples.",
    "2": "Fix patch of #58 (https://github.com/yangheng95/PyABSA/issues/58)"
  },
  "1.0.6": {
    "1": "Fix potential error loading GloVe-based model's checkpoint."
  },
  "1.0.5": {
    "1": "Fix potential error loading ATEPC checkpoint."
  },
  "1.0.4": {
    "1": "Add checkpoint save options, default to save the state_dict instead the whole model",
    "2": "Update documentation of some examples",
    "3": "Fix a dataset selection problem"
  },
  "1.0.1": {
    "1": "Fix #53"
  },
  "0.9.2.1": {
    "1": "fix path of #49"
  },
  "0.9.2.0": {
    "1": "Add text classification (training & inference) support and SST datasets"
  },
  "0.9.1.0": {
    "1": "Add model type check before retraining",
    "2": "Fix syntax distance measure for ATEPC models"
  },
  "0.9.0.6": {
    "1": "Optimize inference printing",
    "2": "Set default encoding=utf-8",
    "3": "Fix graph assigning for ASGCN",
    "4": "Fix a problem may causing failure while search inference datasets"
  },
  "0.9.0.0": {
    "1": "Add BERT baseline models, not available until full test",
    "2": "Refactor some code to allow add model easier",
    "3": "Add post-training feature: to train based on a pretrained PyABSA model, refer to https://github.com/yangheng95/PyABSA/issues/48",
    "4": "Add batch inference (from file) for ATEPC",
    "5": "Fix a bug while predicting sentiment polarity using ATEPC model, refer to https://github.com/yangheng95/PyABSA/issues/47"
  },
  "0.8.9.4": {
    "1": "Fix the inference of DLCF_DCA model"
  },
  "0.8.9.3": {
    "1": "Refactor some code"
  },
  "0.8.9.3rc1": {
    "1": "Add evaluation for glove-based APC models",
    "2": "fix some problems"
  },
  "0.8.9.3rc0": {
    "1": "Add DLCF_DCA_BERT models"
  },
  "0.8.9.2": {
    "1": "Refactor parameter loading method to manage parameters depend on specific model (Note you need to clone the latest examples after updating)",
    "2": "Fix cross validation",
    "3": "Plan to enable BERT for baseline models"
  },
  "0.8.9.1": {
    "1": "Add GloVe models support for APC, available model list: AOA, ASGCN, ATAE-LSTM, Cabasc, IAN, LSTM, MemNet, MGAN, RAM, TC/TD-LSTM, TNet_LF",
    "2": "Add GloVe embedding download support",
    "3": "Next Step: Add inference support for GloVe-based APC models",
    "4": "Please feel free to contribute"
  },
  "0.8.8.8": {
    "1": "Add checkpoint verification",
    "2": "Add release note with open source code",
    "3": "fix param search function"
  },
  "0.8.8.7": {
    "1": "Add release note",
    "2": "Remove some duplicated code"
  },
  "0.8.8.5": {
    "1": "Add new datasets (T-shirt, Television, Copyright belongs to https://github.com/rajdeep345/ABSA-Reproducibility)",
    "2": "Add polarity label-fix features for some datasets containing negative labels",
    "3": "Some typo-fix"
  }
}