ARElight 0.24.0
📘 The complete list of changes
Simplified usability / Flexibility
- 🔧 SampleRowProvider -- Enable Crop Mode #125
- input_data for the pipeline -- design mechanism for supporting an output independing of the sequence of elements #90
- Refactoring installation process and usage as library #61
- D3JS templates and the related pipeline provider #77
- 🔧 Fix launch of the tranformer-based models #78
- ❌ Drop training support (43cac8c)
- Make home folder for states downloading #95
- tqdm -- for inference we may use defined progress bar since we know the total amount of rows #96
- Stemmer -- provide paramer options that allow to enable/disable stemming #97
- googletrans -- Integrate Text Translator (transfer learning) ✨ #108
- Torch and Huggingface
- infer_bert -- switch to LM teminology, support pre-trained states as parameters #70
- ❓ Support of OpenNRE with torch and transformers [NIVTS project findinds] #63
- Remove DeepPavlov and switch to the transformers #69
- Library
- Make it even simpler -- a single cmd script for everything #71
- 🔧 Switch to AREkit==0.24.0
- 🔧 Support new DeepPavlov==1.2.0
- 📦 Make it as a package
- DATA_DIR dependencies -- refactoring #84
- Remove Downloads #85
Minor fixes
- Remove data and predefined text dir #101
- Entities Formatters -- simplify structure, place them in the same infer.py script ✨ #103
- DataFrame processing -- text column customization via cmd parameter ✨ #107
- BERT -- pretrained state might be None if checkpoint is cached. ✨ #106
- Entities Formatters -- simplify structure, place them in the same infer.py script ✨ #103
- Remove pandas dependency #117
- graphs_operations_weighted results in empty graph (no nodes) #112
- Generalization -- settings for label provider/mapping [OpenNRE inference] #119
- from-files -- use DirectoryFilesDocOperation #121
- Setup the correct dependency from the OpenNRE 0.1.1 (ca76a3e)
Change Log
Full Changelog: 0.23.1...0.24.0