=============================
Features
--------
- Add feature_summarizer to produce statistics about the data after
intent resolving to show the users why such decisions are made. (data-summarization)
- Foreshadow is able to run end-to-end with level 1 optimization with the tpot
auto-estimator. (level1-optimization)
- Add Feature Reducer as a passthrough transformation step. (pass-through-feature-reducer)
- Multiprocessing:
1. Enable multiprocessing on the dataset.
2. Collect changes from each process and update the original columnsharer. (process-safe-columnsharer)
- Serialization and deserialization:
1. Serialization of the foreshadow object in a non-verbose format.
2. Deserialization of the foreshadow object. (serialization)
- Adding two major components:
1. usage of metrics for any statistic computation
2. changing functionality of wrapping sklearn transformers to give them DataFrame capabilities. This now uses classes and metaclasses, which should be easier to maintain (74)
- Adding ColumnSharer, a lightweight wrapper for a dictionary that functions
as a cache system, to be used to pass information in the foreshadow pipeline. (79)
- Creating DataPreparer to handle data preprocessing. Data Cleaning is the
first step in this process. (93)
- Adds skip resolve functionality to SmartTransformer, restructure utils, and add is_wrapped to utils (95)
- Add serializer mixin and resture package import locations. (96)
- Add configuration file parser. (99)
- Add Feature Engineerer as a passthrough transformation step. (112)
- Add Intent Mapper and Metric wrapper features. (113)
- Add Preprocessor step to DataPreparer (118)
- Create V2 architecture shift. (162)