tour to horizontal layout, update figures and documentation.

DESCRIPTION

@@ -1,14 +1,13 @@
 Package: cheem
 Title: Interactively Explore Local Explanations with the Radial Tour
-Version: 0.3.1.9000
+Version: 0.4.0.0
 Authors@R: c(
   person("Nicholas", "Spyrison", role = c("aut", "cre"),
     email = "[email protected]",
     comment = c(ORCID = "https://orcid.org/0000-0002-8417-0212"))
   )
-Description: Given a tree-based machine learning model, calculate the tree SHAP 
-    <arXiv:1802.03888>; <https://github.com/ModelOriented/treeshap> 
-    local explanation of every observation. View the data space, explanation 
+Description: Given a non-linear model, calculate the local explanation. 
+    We purpose view the data space, explanation 
     space, and model residuals as ensemble graphic interactive on a shiny 
     application. After an observation of interest is identified, the normalized 
     variable importance of the local explanation is used as a 1D projection

NEWS.md

@@ -1,61 +1,65 @@
-# cheem v0.3.1
-
-- Repaired packagedown site!
-- Fixed news on packagedown site.
-- Recreate the classification saved cases, they fit too well to work for illustrations.
-- Trying to include more axis context in `global_view()`. Will explore text and facet panel titles.
-- Minor documentation and code clean up and clarifications.
-
-
-# cheem v0.3.0 -- Generalized for any attribution
-
-- Rebase all functions from expecting a unified `treeshap::shap()` to generalized
-data frame or matrix format for arbitrary attribution spaces.
-- Rework vignette and examples to reflect this change.
-- Added precomputed predictions and attributions for the Ames, Chocolates, and Penguins datasets. This allows users to run attribution-agnostic functions without dependencies.
-- Add `subset_cheem()`, a convenience function for subsetting cheem lists after construction.
-- Removed plotly subplot variations of visuals: `global_view_subplots()`, `radial_cheem_tour_subplots()`. These were development variations never used in the shiny app.
-- Minor function renames for parsimony and consistency.
-
-
-# cheem v0.2.0 (CRAN releases here on out)
-
-## App related changes
-
-- Added vignette: _Getting started with cheem_.
-- Added __pkgdown__ site: https://nspyrison.github.io/cheem/.
-- Added global model performance metrics to shiny app.
-- In `global_view()`, added yhaty panel (residual plot/confusion matrix).
-- In `global_view()`, added color options: log_maha.data and cor_attr.y.
-- In `cheem_radial_tour()`, added regression case panel with additional fixed y of residual.
-- In app radial tour inputs, added inclusion variable, subsetting variables used in radial tour.
-- `plotly::subplot()` variants of `global_view()` & `cheem_radial_tour()`. 
-- Added AmesHousing data, chocolates, and new toy simulated datasets (shiny app only).
-- Reduced shiny app wording.
-
-## Internal & utilities
-
-- Major rebase of `cheem_ls()`.
-- Added `linear/logistic_tform()` to suggest an alpha as a function of the number of observations.
-
-
-## Sourcing __treeshap__
-
-- __drat__ repository hosting __treeshap__ did not work with debian and window rhub platforms;
-- Minimally ported functions and cpp source files with @author & @source. Changed examples for consistency and smaller code base support.
-- as of v0.3.0, cheem was generalized to all local variable attributions, so this is not a concern.
-
-
-# cheem v0.1.0 (GitHub only, commit 283da4)
-
-## Primary preprocessing functions
-
-- `default_rf()` create a `randomForest::randomForest()` with more conservative defaults.
-- `attr_df_treeshap()` create `treeshap::treeshap()` local explanations of each observation.
-- `cheem_ls()` create a cheem list of prepared tables for use in `run_app()`.
-
-## Primary visual functions
-
-- `run_app()` which is a shiny app consuming the following two:
-- `global_view()` linked 'plotly' of approximations of data- and attribution-spaces with model information. 
-- `cheem_radial_tour()` create `spinifex::ggtour` of the specified radial tour. Consumed by animate_plotly, animate_gganimate, or filmstrip.
+# cheem v0.4.0
+
+- Repaired packagedown site!
+- Fixed news on packagedown site.
+- Shiny app has go buttons rather than waiting after every input change.
+- Shiny app text, plot dimensions, and text cleaned up.
+- Classification tour now uses a horizontal layout.
+- Cleaned up the text on the facet panels for `global_tour()` and `radial_cheem_tour()`.
+- Recreate the saved classification model, they fit too well to work as illustrations.
+- Set seed more consistently. All model and attribution shifted a bit, but will be more replicable going forward.
+- Minor documentation and code clean up and clarifications.
+
+
+# cheem v0.3.0 -- Generalized for any attribution
+
+- Rebase all functions from expecting a unified `treeshap::shap()` to generalized
+data frame or matrix format for arbitrary attribution spaces.
+- Rework vignette and examples to reflect this change.
+- Added precomputed predictions and attributions for the Ames, Chocolates, and Penguins datasets. This allows users to run attribution-agnostic functions without dependencies.
+- Add `subset_cheem()`, a convenience function for subsetting cheem lists after construction.
+- Removed plotly subplot variations of visuals: `global_view_subplots()`, `radial_cheem_tour_subplots()`. These were development variations never used in the shiny app.
+- Minor function renames for parsimony and consistency.
+
+
+# cheem v0.2.0 (CRAN releases here on out)
+
+## App related changes
+
+- Added vignette: _Getting started with cheem_.
+- Added __pkgdown__ site: https://nspyrison.github.io/cheem/.
+- Added global model performance metrics to shiny app.
+- In `global_view()`, added yhaty panel (residual plot/confusion matrix).
+- In `global_view()`, added color options: log_maha.data and cor_attr.y.
+- In `cheem_radial_tour()`, added regression case panel with additional fixed y of residual.
+- In app radial tour inputs, added inclusion variable, subsetting variables used in radial tour.
+- `plotly::subplot()` variants of `global_view()` & `cheem_radial_tour()`. 
+- Added AmesHousing data, chocolates, and new toy simulated datasets (shiny app only).
+- Reduced shiny app wording.
+
+## Internal & utilities
+
+- Major rebase of `cheem_ls()`.
+- Added `linear/logistic_tform()` to suggest an alpha as a function of the number of observations.
+
+
+## Sourcing __treeshap__
+
+- __drat__ repository hosting __treeshap__ did not work with debian and window rhub platforms;
+- Minimally ported functions and cpp source files with @author & @source. Changed examples for consistency and smaller code base support.
+- as of v0.3.0, cheem was generalized to all local variable attributions, so this is not a concern.
+
+
+# cheem v0.1.0 (GitHub only, commit 283da4)
+
+## Primary preprocessing functions
+
+- `default_rf()` create a `randomForest::randomForest()` with more conservative defaults.
+- `attr_df_treeshap()` create `treeshap::treeshap()` local explanations of each observation.
+- `cheem_ls()` create a cheem list of prepared tables for use in `run_app()`.
+
+## Primary visual functions
+
+- `run_app()` which is a shiny app consuming the following two:
+- `global_view()` linked 'plotly' of approximations of data- and attribution-spaces with model information. 
+- `cheem_radial_tour()` create `spinifex::ggtour` of the specified radial tour. Consumed by animate_plotly, animate_gganimate, or filmstrip.

R/1_cheem_lists.r

@@ -166,7 +166,7 @@ global_view_df_1layer <- function(
 #'   comp <- 2
 #'   global_view(peng_chm, primary_obs = prim, comparison_obs = comp)
 #'   bas <- sug_basis(peng_xgb_shap, prim, comp)
-#'   mv  <- sug_manip_var(peng_xgb_shap, primary_obs = 1, comparison_obs = 2)
+#'   mv  <- sug_manip_var(peng_xgb_shap, primary_obs = prim, comp)
 #'   ggt <- radial_cheem_tour(peng_chm, basis = bas, manip_var = mv)
 #'   animate_plotly(ggt)
 #' }
@@ -203,7 +203,7 @@ global_view_df_1layer <- function(
 #'   comp <- 2
 #'   global_view(ames_rf_chm, primary_obs = prim, comparison_obs = comp)
 #'   bas <- sug_basis(ames_rf_shap, prim, comp)
-#'   mv  <- sug_manip_var(ames_rf_shap, primary_obs = 1, comparison_obs = 2)
+#'   mv  <- sug_manip_var(ames_rf_shap, primary_obs = prim, comp)
 #'   ggt <- radial_cheem_tour(ames_rf_chm, basis = bas, manip_var = mv)
 #'   animate_plotly(ggt)
 #' }

R/2_visualization.r

@@ -584,26 +584,25 @@ radial_cheem_tour <- function(
   ### Classification case -----
   if(.prob_type == "classification"){
     .pred_clas <- decode_df$predicted_class
+    .facet_fore <- rep("attribution projection", each = .n)
     ## ggtour
     ggt <- spinifex::ggtour(.mt_path, .dat, angle = angle,
                             do_center_frame = do_center_frame) +
+      spinifex::facet_wrap_tour(facet_var = .facet_fore, nrow = 1) +
       ## Density
       spinifex::proto_density(
         aes_args = list(color = .pred_clas, fill = .pred_clas),
         row_index = row_index, rug_shape = pcp_shape) +
-
-      #Warning message:
-      #In Ops.factor(yscale, x[, 2]) : '*' not meaningful for factors
       ## PCP on Basis, 1D
       proto_basis1d_distribution(
         cheem_ls$attr_df,
         primary_obs = .prim_obs, comparison_obs = .comp_obs,
-        position = "bottom1d", group_by = .pred_clas,
+        position = "floor1d", group_by = .pred_clas,
         do_add_pcp_segments = as.logical(do_add_pcp_segments),
         pcp_shape = pcp_shape, inc_var_nms = inc_var_nms,
        row_index = row_index) +
       ## Basis 1D
-      spinifex::proto_basis1d(position = "bottom1d", manip_col = "black") +
+      spinifex::proto_basis1d(position = "floor1d", manip_col = "black") +
       spinifex::proto_origin1d() +
       ## Highlight comparison obs
       spinifex::proto_highlight1d(
@@ -641,7 +640,7 @@ radial_cheem_tour <- function(
       ## Foreground:
       .dat_fore   <- rbind(.dat, .dat)
       .idx_fore   <- c(row_index, row_index)
-      .facet_fore <- factor(rep(c("observed y", "residual"), each = 2 * .n))
+      .facet_fore <- factor(rep(c("attribution projection by observed y", "attribution projection by residual"), each = 2 * .n))
       .fixed_y    <- c(.y, .resid)
     } else {
       ## not doubled up data; just fixed_observed y
@@ -652,7 +651,7 @@ radial_cheem_tour <- function(
       ## Foreground:
       .dat_fore   <- .dat
       .idx_fore   <- row_index
-      .facet_fore <- rep("observed y", each = .n)
+      .facet_fore <- rep("attribution projection by observed y", each = .n)
       .class_fore <- .class
       .fixed_y    <- .y
     }