Hi, I want to use Dalex with a model created from Autkeras, with a self-made Sequential Model it worked. What is the problem?

Model from Autokeras structured_data_classifier : <tensorflow.python.keras.engine.functional.Functional at 0x1bbab227340> https://autokeras.com/structured_data_classifier/

explainer = dx.Explainer(
    model = model,
    data = X_test,
    y = y_test,
    model_type= "classification")

#%%
explainer.model_performance()

explainer.model_parts().plot()

I have the following code, for parameters tuning with mlr3

df=readARFF("xerces.arff") task=TaskRegr$new("df", df, target = "price") learner= lrn("regr.rpart") resampling = rsmp("repeated_cv", folds=3, repeats=3) measure= msr("regr.mae")

search_space = paradox::ParamSet$new( params = list(paradox::ParamDbl$new("cp", lower = 0.001, upper = 0.1))) terminator = trm("evals", n_evals = 15)

tuner = tnr("grid_search")

at = AutoTuner$new( learner = learner, resampling = resampling, measure = measure, search_space = search_space, terminator = terminator, tuner = tuner, store_tuning_instance = TRUE, store_benchmark_result = TRUE, store_models = TRUE )

bmr = benchmark(grid,store_models = TRUE)

My question is how can I create an explainer on this model?

explainer = explain_mlr3( ??,...........)

I have used Break-down method for instance level explanation and does work fine. I have never used LIME method and now when I am using it, it gives me the following error:

Error in [.data.frame(explainer$data, , colnames(new_observation)) : undefined columns selected

My code is:

explainer5 = explain_mlr3(model5, data = test[,-21], y = as.numeric(test$report)-1, label="SVM")

new_observation= test[6,] plot(predict_parts(explainer5, new_observation = new_observation, type = "break_down_interactions")) //// This works fine

/// Problem is in the following code

model_type.dalex_explainer <- DALEXtra::model_type.dalex_explainer predict_model.dalex_explainer <- DALEXtra::predict_model.dalex_explainer

lime_tool <- predict_surrogate(explainer = explainer5, new_observation = new_observation, n_features = 3, n_permutations = 1000, type = "lime")

Error in [.data.frame(explainer$data, , colnames(new_observation)) : undefined columns selected

What could be the problem? I am taking help from the example in https://ema.drwhy.ai/LIME.html

Hi I have a bit confusion over a plot of instance-level explanation.. I have to predict something which has a value of 0 to 10, while 10 means a product is very expensive and 0 means it is not expensive at all. I get a prediction of 0.7 (which of course means not expensive) and majority of input features have negative impact (red bars instead of green bar). So my question is how to interpret this explanation. If the value of this input feature will decreases, the prediction will decrease (go from 0.7 to 0.0) or the prediction will increase (goes from 0.7 to higher values such as 10).

Thank you and hope you have understood my point

I've noticed different results of feature_importance in DALEX and ingredients. It is caused by different default values for loss_function. I was wondering shouldn't they be the same for consistency reasons.

https://github.com/ModelOriented/DALEX/blob/8858926c4abb5b7cd7d1a16bfd9f6df2434003b0/R/model_parts.R#L43

https://github.com/ModelOriented/ingredients/blob/9f8f82f05ddb87e4abf2fb20846f097113522a33/R/feature_importance.R#L152

> library("DALEX")
> model_titanic_glm <- glm(survived ~ gender + age + fare,
                         data = titanic_imputed, family = "binomial")
> explain_titanic_glm <- explain(model_titanic_glm,
                               data = titanic_imputed[,-8],
                               y = titanic_imputed[,8])

> DALEX::feature_importance(explain_titanic_glm, n_sample = NULL) 
      variable mean_dropout_loss label
1 _full_model_          370.1496    lm
2        class          370.1496    lm
3     embarked          370.1496    lm
4        sibsp          370.1496    lm
5        parch          370.1496    lm
6          age          371.6003    lm
7         fare          384.4722    lm
8       gender          548.2130    lm
9   _baseline_          599.7132    lm

> ingredients::feature_importance(explain_titanic_glm)
      variable mean_dropout_loss label
1 _full_model_         0.4095316    lm
2        class         0.4095316    lm
3     embarked         0.4095316    lm
4        sibsp         0.4095316    lm
5        parch         0.4095316    lm
6          age         0.4101522    lm
7         fare         0.4173782    lm
8       gender         0.4976695    lm
9   _baseline_         0.5178448    lm

> ingredients::feature_importance(explain_titanic_glm, loss_function = DALEX::loss_sum_of_squares)
      variable mean_dropout_loss label
1 _full_model_          370.1496    lm
2        class          370.1496    lm
3     embarked          370.1496    lm
4        sibsp          370.1496    lm
5        parch          370.1496    lm
6          age          371.0894    lm
7         fare          384.0622    lm
8       gender          546.5614    lm
9   _baseline_          598.0323    lm

(Because of https://github.com/ModelOriented/DALEX/issues/175, n_sample = NULL in DALEX::feature_importance)

Hello

I want to evaluate the partial dependence profile of a variable with RF model (with default settings) and RF (with optimized settings). Can I do it?

default_model=model_profile(explainer = my_explainer, variables = "var1") opt_model=model_profile(explainer = my_explainer, variables = "var1")

I trained 5 models with caret::train. The following code works for svmRadial, gbm and rf, BUT NOT for lm, knn:

DALEX.explainer <- model %>%
  DALEX::explain(
    data = features,
    y = model$trainingData$.outcome >= UPPER.MIN,
    label = paste(model$method, " model"),
    colorize = TRUE
  )

DALEX.attribution <- DALEX.explainer %>%
  iBreakDown::local_attributions(
    local.obs,
    keep_distributions = TRUE
  )

throws error:

Error in `[.data.frame`(newdata, , object$method$center, drop = FALSE) : 
  undefined columns selected

In the documentation, I couldn't find any hint that the function is model-dependent. Any clue why this doesn't work for lm and knn??

Im using DALEX in R, trying to create a break_down plot from a new observation and a XGBoost model.

I succesfully created the explain object and the predict_parts break_down dataframe, which BTW took ages (more than a day to calculate - anyone have an intuitive idea why it takes so long or how to speed things up?).

But plotting the predict_parts only gives a scatterplot, I was expecting a nice waterfall-looking plot.

Im sorry for not being able to post a minimal reproducible example and not being able to share data or model, but here ist the snippet:

library(DALEX)
  
explain <- DALEX::explain(trained_model, data = train, y = as.numeric(train$PRUEFERGEBNIS)-1, type = "classification", label = "xgboost", predict_function = 
                              function(trained_model, test){
                                previous_na_action <- options('na.action')
                                options(na.action='na.pass')
                                sparse_matrix_test <- sparse.model.matrix(PRUEFERGEBNIS ~.
                                                                          , data = test)
                                options(na.action=previous_na_action$na.action)
                                
                                results_test <- predict(trained_model, sparse_matrix_test, type = "response")
                                round(results_test,3)
                              })

parts_plot <- predict_parts(explain, new_observation = test[1, , drop = FALSE], type = "break_down")
plot(parts_plot)

Wonderful package, thanks a lot, I am using it quite frequently.

Here are some suggestions, especially having models with case weights in mind:

1. The function model_performance allows to pass a loss function of the residuals (as single argument) only. This seems too restrictive as many loss functions cannot be written as a function of residuals but rather of the observed value and the predicted one. Most DALEX functions allow a general loss function with two arguments, not just one. This inconsitency is e.g. a problem when we want to plot deviance residuals of a logistic regression.

2. Ability to enhance plots with ggplot functions, i.e. allowing to add e.g. + geom_points(...)

3. variable_importance: In our case, the custom predict function depends on a weight vector w (basically the raw predictions are multiplied by w). Since the DALEX interface requires this vector to be a column in the data frame passed to explain, w currently appears in the variable_importance output as well. Would it be an option to add an optional "dropVars = NULL" argument to variable_importance (or even in explain)?

4. Like issue 3, but for the single_prediction part. Here, it is not only an aesthetic problem if a "special" column like case weight w appears.

Did the prescribed: devtools::install_github("pbiecek/DALEX") but Rstudio (latest version under Linux), will not complete the DALEX install.

After the above command, R will just hang and nothing happens... R will just hang there.

Waited for 5 minutes, then suspended the installation.

What am I missing in order to install DALEX ?. (all other R packages have installed with no problems...).

Thanks!

Hi

If I have 100 observations/instances and want to create the PD profile of few important features, then what is the difference between the PD profile of these instances using

variab= c("var1","var2", "var3" ) pdp <- model_profile(explainer = explainer, variables = variab) plot(pdp)

and between the CP profile when we manually specify the instances like

new_observation= data[c(2,7,8,11,15,16,21,24,25,26,30,45,46),] cp<- ceteris_paribus(explainer, new_observation) cp_agg <- aggregate_profiles(cp, variables = variab) plot(cp_agg)

I would like to use loss_accuracy as my loss function in model_parts(), but whenever I use it, the mean_drop_loss is always 0. I have tried loss_accuracy for regression, classification, and multiclass classification (see reprex below). Am I using it correctly?

library(DALEX)
#> Welcome to DALEX (version: 2.4.2).
#> Find examples and detailed introduction at: http://ema.drwhy.ai/
library(ranger)
df <- mtcars[, c('mpg', 'cyl', 'disp', 'hp', 'vs')]
# Regression
reg <- lm(mpg ~ ., data = df)
explainer_reg <- explain(reg, data = df[,-1], y = df[,1])
#> Preparation of a new explainer is initiated
#>   -> model label       :  lm  (  default  )
#>   -> data              :  32  rows  4  cols 
#>   -> target variable   :  32  values 
#>   -> predict function  :  yhat.lm  will be used (  default  )
#>   -> predicted values  :  No value for predict function target column. (  default  )
#>   -> model_info        :  package stats , ver. 4.2.2 , task regression (  default  ) 
#>   -> predicted values  :  numerical, min =  12.56206 , mean =  20.09062 , max =  27.04625  
#>   -> residual function :  difference between y and yhat (  default  )
#>   -> residuals         :  numerical, min =  -4.019038 , mean =  1.010303e-14 , max =  6.976988  
#>   A new explainer has been created!
feature_importance(explainer_reg, loss_function = loss_accuracy)
#>       variable mean_dropout_loss label
#> 1 _full_model_                 0    lm
#> 2          cyl                 0    lm
#> 3         disp                 0    lm
#> 4           hp                 0    lm
#> 5           vs                 0    lm
#> 6   _baseline_                 0    lm
# Classification
classif <- glm(vs ~ ., data = df, family = binomial)
explainer_classif <- explain(classif, data = df[,-5], y = df[,5])
#> Preparation of a new explainer is initiated
#>   -> model label       :  lm  (  default  )
#>   -> data              :  32  rows  4  cols 
#>   -> target variable   :  32  values 
#>   -> predict function  :  yhat.glm  will be used (  default  )
#>   -> predicted values  :  No value for predict function target column. (  default  )
#>   -> model_info        :  package stats , ver. 4.2.2 , task classification (  default  ) 
#>   -> predicted values  :  numerical, min =  7.696047e-06 , mean =  0.4375 , max =  0.9920295  
#>   -> residual function :  difference between y and yhat (  default  )
#>   -> residuals         :  numerical, min =  -0.9474062 , mean =  -1.483608e-12 , max =  0.5318376  
#>   A new explainer has been created!
feature_importance(explainer_classif, loss_function = loss_accuracy)
#>       variable mean_dropout_loss label
#> 1 _full_model_                 0    lm
#> 2          mpg                 0    lm
#> 3          cyl                 0    lm
#> 4         disp                 0    lm
#> 5           hp                 0    lm
#> 6   _baseline_                 0    lm
# Multiclass classification
multiclass <- ranger(cyl ~ ., data = df, probability = TRUE)
explainer_multiclass <- explain(multiclass, data = df[,-2], y = df[,2])
#> Preparation of a new explainer is initiated
#>   -> model label       :  ranger  (  default  )
#>   -> data              :  32  rows  4  cols 
#>   -> target variable   :  32  values 
#>   -> predict function  :  yhat.ranger  will be used (  default  )
#>   -> predicted values  :  No value for predict function target column. (  default  )
#>   -> model_info        :  package ranger , ver. 0.14.1 , task multiclass (  default  ) 
#>   -> model_info        :  Model info detected multiclass task but 'y' is a numeric .  (  WARNING  )
#>   -> model_info        :  By deafult multiclass tasks supports only factor 'y' parameter. 
#>   -> model_info        :  Consider changing to a factor vector with true class names.
#>   -> model_info        :  Otherwise I will not be able to calculate residuals or loss function.
#>   -> predicted values  :  predict function returns multiple columns:  3  (  default  ) 
#>   -> residual function :  difference between 1 and probability of true class (  default  )
#>   -> residuals         :  the residual_function returns an error when executed (  WARNING  ) 
#>   A new explainer has been created!
feature_importance(explainer_multiclass, loss_function = loss_accuracy)
#>       variable mean_dropout_loss  label
#> 1 _full_model_                 0 ranger
#> 2          mpg                 0 ranger
#> 3         disp                 0 ranger
#> 4           hp                 0 ranger
#> 5           vs                 0 ranger
#> 6   _baseline_                 0 ranger

^{Created on 2022-12-29 with reprex v2.0.2}

When I try other loss functions (e.g., loss_root_mean_square for regression, loss_one_minus_auc for classification), they return non-zero values.

feature_importance(explainer_reg, loss_function = loss_root_mean_square)
#>       variable mean_dropout_loss label
#> 1 _full_model_          2.844520    lm
#> 2           vs          2.861546    lm
#> 3           hp          3.328176    lm
#> 4         disp          4.201312    lm
#> 5          cyl          4.498485    lm
#> 6   _baseline_          7.777811    lm
feature_importance(explainer_classif, loss_function = loss_one_minus_auc)
#>       variable mean_dropout_loss label
#> 1 _full_model_        0.03571429    lm
#> 2          mpg        0.04603175    lm
#> 3         disp        0.04642857    lm
#> 4           hp        0.31785714    lm
#> 5          cyl        0.36031746    lm
#> 6   _baseline_        0.51884921    lm

^{Created on 2022-12-29 with reprex v2.0.2}

Is there something different about how loss_accuracy is used?

I'm using DALEX v2.4.2, R v4.2.2, RStudio v2022.12.0+353, Ubuntu 22.04.1

Hi, I am using your fairness package for my project, and I wondered about your fairness checks:

For the test equal opportunity ratio, you use TPR, not FNR (the True ratio and not the complementary False ratio). but, For the test "predictive equality ratio" you use FPR, not TNR (the False ratio, not the complementary True ratio).

I wondered what is the justification for TPR and FPR specifically (and not TPR and TNR for example). Thanks!

I use DALEX in most of my ML research projects. I keep getting a criticism from reviewers regarding a drawback of permutation based method in the presence of multiple correlated predictors. They argue that if there is a group of highly correlated but important predictors, they may not show up at the top of feature importance. Can someone comment on this criticism? How is this addressed in DALEX?

I'm trying to wrap a fastai tabular learner with DALEX. I got 'DataFrame' object has no attribute 'to_frame' error with dx.Explainer(learn, xs, y, label = "Deep NN"). Any potential problems with this line of code? Thanks!

In R package show_boxplots = TRUE would plot the permutations in feature importance In python model_parts, this parameter does not seem to exist. Or it is called something else?