Changes in this release:
- Internal structure of `CondDensityEstimator` now matches that of the `CauseClusterer` and `EffectClusterer`, where each of these Block types takes a "model" object as input.
- Abstract classes `CDEModel` and `ClustererModel` have been added to enforce model interfaces that the respective block types expect. `adding_blocks.ipynb` provides an example for specifying new models.
- Hyperparameters for blocks and models are now separated by specifying model params as a nested dictionary in the block param dictionary
- User interface for Clusterer tuning has been polished
- Verbosity behavior has been resolved so that all user set flags are followed
- Tests have been updated and reorganized
v.1.1.0
Changes in this release:
- `post_cfl` suite of analyses to run on a trained `Experiment`, providing a more intuitive understanding of the results CFL generates. This includes
- intervention recommendation
- feature importance computation
- Clearer module naming
- Updated documentation examples now include automated clusterer training and most recent visualizations
- Updated docs and tests for new naming scheme
- a `CondExpDIY` model for users who want to define their own model architecture in keras instead of through `CondExpMod` params
- Basic contributor documentation
- pep8 style formatting
v.1.0.3
This patch:
- updates some tutorial notebooks to match the new cfl interface
- makes cfl logging more transparent during training and prediction
v.1.0.2
1. This release adds basic functionality and documentation for visualization macrostates as averages of their microvariable constituents. The added module can be found at `cfl/visualization_methods/basic_visualizations.py`, and the corresponding documentation is [here](https://cfl.readthedocs.io/en/latest/examples/basic_visualizations.html).
2. The `Clusterer` block-type in the previous version of the code has been split into two blocks: `CauseClusterer` and `EffectClusterer` for the sake of conceptual clarity and ease of parameter tuning. CFL examples have been updated to demonstrate how to parameterize these new types.