DALEXverse 0.19.8 release summer 2019 about ingredients

[hbaniecki]

-1. readability: DESCRIPTION change URL to https://github.com/ModelOriented/ingredients
update demo.png in README

0. in many places in documentation x should be documented as explainer first and
   data/y/predict_function should be placed as last parameters in function
   ex: conditional_dependency.R, feature_importance.R, ...

1. placement of ... parameter is inconsistent across many functions

• accumulated_dependency.R
• aspect_importance.R
• conditional_dependency.R
• feature_importance.R
• partial_dependency.R
• more

2. parameters in documentation are in different order than in function call

• aggregate_profiles.R
• cluster_profiles.R

3. change parameter name

• aspect_importance.R aspects_list could be just aspects (like variables)
• plot_ceteris_paribus_2d_explainer.R split_ncol to facet_ncol

4. In many functions N and B are beeing used. There is no consistency

5. check x class and produce error in calculate_oscillations.R could also appear in

• aggregate_profiles.R
• cluster_profiles.R
• plot_aggregated_profiles_explainer.R
• plot_ceteris_paribus_2d_explainer.R
• plot_ceteris_paribus_explainer.R
• more

6. = instead of <-

• https://github.com/ModelOriented/ingredients/blob/master/R/calculate_oscillations.R#L61
• https://github.com/ModelOriented/ingredients/blob/master/R/ceteris_paribus_2d.R#L88
• https://github.com/ModelOriented/ingredients/blob/master/R/cluster_profiles.R#L138
• https://github.com/ModelOriented/ingredients/blob/master/R/aspect_importance.R#L171
• https://github.com/ModelOriented/ingredients/blob/master/R/ceteris_paribus.R#L167
• https://github.com/ModelOriented/ingredients/blob/master/R/describe_aggregatedprofiles.R#L74
• https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L102
• https://github.com/ModelOriented/ingredients/blob/master/R/describe_featureimportance.R#L56
• https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L312
• https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L338 (all sufix =)

7. CODE NOTES

• ifelse https://github.com/ModelOriented/ingredients/blob/master/R/describe_aggregatedprofiles.R#L267
• -> https://github.com/ModelOriented/ingredients/blob/master/R/calculate_oscillations.R#L53
• could use some comments https://github.com/ModelOriented/ingredients/blob/master/R/ceteris_paribus_2d.R
• could use some comments https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L307
• could use some comments https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L321
• this is prone to ERROR - needs to be changed https://github.com/ModelOriented/ingredients/blob/master/R/conditional_dependency.R#L84
  and https://github.com/ModelOriented/ingredients/blob/master/R/conditional_dependency.R#L85
• take care of class logical (?) https://github.com/ModelOriented/ingredients/blob/master/R/describe_aggregatedprofiles.R#L56
  https://github.com/ModelOriented/ingredients/blob/master/R/describe_ceterisparibus.R#L82
• add predict as default https://github.com/ModelOriented/ingredients/blob/master/R/feature_importance.R#L162
• this could be default parameter value in function call https://github.com/ModelOriented/ingredients/blob/master/R/plotD3_aggregated_profiles_explainer.R#L150 the same in other plotD3

8. calculate_variable_profile.R has no support for DALEX explainer use + sad documentation

9. ceteris_paribus_2d.R all parameters are inconsistent with the rest of ingredients

10. describe.R could be deleted?

11. Change file name:

• describe_aggregatedprofiles.R

• describe_ceterisparibus.R

• describe_featureimportance.R

• plotD3_aggregated_profiles_explainer.R

• plotD3_ceteris_paribus_explainer.R

• plotD3_feature_importance_explainer.R

• plot_aggregated_profiles_explainer.R

• and more..... (cut _explainer everywhere)

• cut s: print_ceteris_paribus_descriptions.R

12. weird/no parameter documentation

• describe_aggregatedprofiles.R
• describe_ceterisparibus.R
• describe_featureimportance.R
  https://github.com/ModelOriented/ingredients/blob/master/R/select_samples.R#L11

13. ingredients has theme_drwhy.R

14. add @noRd to not exported functions

[AdamIzdebski]

(Key functions)
readability: documentation (title, description, details)

• feature_importance()
  1. Make descriptions consistent with https://pbiecek.github.io/PM_VEE/
     possible change: "This function calculates permutation based feature importance."
  2. Arguments:
     x should the description include model or just class explainer (Argument in favour of deleting the model option: function feature_importance needs arguments that are additional and are on the low end of the list in order to work with model.)
     typos,
     'type' possible change "a string specifying which ... . Possible values are ... "
     punctuation is not consistent it's either a comma or dot after argument class and before the description
• ceteris_paribus()
  3. description: The first sentence should communicate what the function does. possible change: "This function calculates a one dimensional ceteris paribus profile for the selected variables."
  4. Arguments order is not consistent, new_observation is needed for the function to work and should be second in order.
  new_observation should the observation contain the decision (y) variable?
• calculate_oscillations() 2 applies
• ceteris_paribus_2d() description that could serve as an example for other functions, arguments 'variables' in not clear how to use
• partial_dependency() 2 applies,
• conditional_dependency() 1,3`

readability: examples (relevant, complete, with comments)

• feature_importance() if x can be a model, then an example how to use the function with models should be added
• ceteris_paribus() 5. examples could be commented, more different arguments could be demonstrated on the same explainer
• ceteris_paribus_2d() 6. example for using variables argument should be provided, it is not always working
• partial_dependency() 5 and 6 applies
• conditional_dependency() 5,6
• aggregate_profiles() warnings are generated, 6

reproducibility: tests (code coverage)

• describe_aggregated_profiles.R - test on inputs with continuous variable
• aggregated_profiles.R - test accumulated profiles on non null groups and non-numeric variables, test conditional_profiles for nonnumeric variables, those examples can be ploted too.

links to functions: \code
All functions need link to DALEX::explain in arguments
argument x linked to ceteris_paribus() in aggregate_profiles() , accumulated_dependency()

[maksymiuks]

Try to get rid of titanic package suggest

[hbaniecki]

add demo gif to ingredients and iBreakDown