Added
- Formal documentation hosted on readthedocs.org
- Keep trace of historical documentations
- Unit tests
- `info_plots.target_plot_interact`: visualise average target value across interaction between two features
- `info_plots.actual_plot_interact`: visualise prediction distribution across interaction between two features
- `get_dataset`: store models and datasets for three different problems
(binary classification, multi-class classification, regression)
- Tutorials in jupyter notebook format
Changed
- Move all information related plots under `info_plots`, including
- `info_plots.target_plot`
- `info_plots.target_plot_interact`
- `info_plots.actual_plot`
- `info_plots.actual_plot_interact`
- Move all utility functions under `xx_utils.py`
- `utils.py`: general utility functions
- `info_plot_utils.py`: utility functions for information plots
- `pdp_calc_utils.py`: utility functions for pdp related calculation
- `pdp_plot_utils.py`: utility functions for pdp related plots
- `class PDPIsolate`
- Rename `class pdp_isolate_obj` as `class PDPIsolate`
- Remove `self.classifier`, `self.model_features`, `self.actual_columns`: useless
- Add `self.which_class`, `self.percentile_info`, `self.count_data`, `self.hist_data`:
store class information for multi-class problem,
store percentile information for grid points,
store value count information as well as feature values for numeric feature
- `class PDPInteract`
- Rename `class pdp_interact_obj` as `class PDPInteract`
- Remove `self.classifier`, `self.model_features`: useless
- Add `self.which_class`: store class information for multi-class problem
- Combine `self.pdp_isolate_out1` and `self.pdp_isolate_out2` into `self.pdp_isolate_outs`
- `pdp.pdp_isolate`
- Replace `train_X` as `dataset` to store whole dataset
instead of only the subset for model training,
thus add `model_features` to indicate features used for model training
- Add `grid_type`, `grid_range`: define type and range for grid points
- Add `memory_limit`, `n_jobs`: limit memory usage, support parallel processing
- Set `predict_kwds` default value into `None` instead of `{}`
- Add `data_transformer`: support dataset transformation
- `pdp.pdp_plot`
- Add `plot_pts_dist`: enable to plot distribution of data points
- Remove `plot_org_pts`: no longer support plotting original data points
- Set `cluster_method` default value as 'accurate' instead of None
- Add `show_percentile`: show percentile information of grid points
- Set `ncols` default value as 2 instead of None
- Add `which_classes`, remove `multi_flag`, `which_class`:
plot for a single class is now supported by `which_classes`
- `pdp.pdp_interact`
- Replace `train_X` as `dataset` to store whole dataset
instead of only the subset for model training,
thus add `model_features` to indicate features used for model training
- Set `num_grid_points` default value as None instead of `[10, 10]`
- Add `grid_type`, `grid_range`: define type and range for grid points
- Set `percentile_ranges` default value as None instead of `[None, None]`
- Set `cust_grid_points` default value as None instead of `[None, None]`
- Set `predict_kwds` default value into `None` instead of `{}`
- `pdp.pdp_interact_plot`
- Add `plot_type`, `plot_pdp`, remove `only_inter`: define plot type and whether to plot pdp
for both features, only showing contour plot now is supported by `plot_type` and `plot_pdp`
- Add `which_classes`, remove `multi_flag`, `which_class`:
plot for a single class is now supported by `which_classes`
- Set `ncols` default value as 2 instead of None
- Remove `center`, `plot_org_pts`, `plot_lines`, `frac_to_plot`, `cluster`, `n_cluster_centers`,
`cluster_method`: no longer support plotting separate pdp plots
- `info_plots.target_plot`
- Add `grid_type`, `grid_range`: define type and range for grid points
- Add `show_percentile`: show percentile information of grid points
- Add `show_outliers`: whether to show data points outside the grid range
- Add `endpoint`: whether stop is the last grid point
- Add `ncols`: define number of columns for multiple plots
- `info_plots.actual_plot`
- Add `model`, `X`, `feature`, remove `pdp_isolate_out`: no longer depend on `pdp.pdp_isolate`,
thus need to define all necessary parameters for calculating the results
- Add `num_grid_points`, `grid_type`, `percentile_range`, `grid_range`, `cust_grid_points`,
`show_percentile`, `show_outliers`, `endpoint`, `which_classes`, `predict_kwds`
- Set `ncols` default value as 2 instead of None
- Add `which_classes`, remove `multi_flag`, `which_class`:
plot for a single class is now supported by `which_classes`
- Set `predict_kwds` default value into `None` instead of `{}`
Fixed
- Python3 compatibility
- All plotting related functions would return a `matplotlib.figure.Figure` object
as well as `Matplotlib.axes` for further modification