Strong Improvements
Pre-release
Pre-release
supported features
- v0.0.1: computation of co-occurrence matrix
- v0.0.1: in-library encoding using pre-trained models
- v0.0.3: search using tags that are not available in tag_list
new features
- v0.0.4: int8 quantization
- v0.0.4: more intuitive validation technique
- v0.0.4: added "dot_product" and "PCA" options
deprecated features
- v0.0.2: in-library compression/expansion of the one_hot vector
known issues
- length of the covariate vector can only be equivalent to the length of the ones in df_M
- does not support a sparse vector data structure, necessary for highly scalable scenarios
- search using custom tags is only available when vectors are encoded using the pre-trained encoder
- no code to handle the expansion of the existing tag_list
- no methods to optimize dot_product calculation
- no clustering methods to maintain constant the size of tag_list
- code for encrypted covariate search exists but has not yet been implemented in the library
- code for covariate tagging exists but has not yet been implemented in the library
- search methods still maintain the old name "tag_filtering", and need to be renamed as "search"