16.9 LIME

Local Interpretable Model-agnostic Explanations (LIME) is an algorithm that helps explain individual predictions which assumes that every complex model is linear on a local scale after applying the next steps:

  1. Permute your training data to create new data points that are similar to the original one, but with some changes in the feature values.
  2. Compute proximity measure (e.g., 1 - distance) between the observation of interest and each of the permuted observations.
  3. Apply selected machine learning model to predict outcomes of permuted data.
  4. Select \(m\) number of features to best describe predicted outcomes.
  5. Fit a simple model (LASSO) to the permuted data, explaining the complex model outcome with \(m\) features from the permuted data weighted by its similarity to the original observation.
  6. Use the resulting feature weights to explain local behavior.

Source: https://ema.drwhy.ai/LIME.html

16.9.1 Implementation

In R the algorithm has been written in iml, lime and DALEX. Let’s use the iml one:

  1. Create a list that contains the fitted machine learning model and the feature distributions for the training data.
# Create explainer object
components_lime <- lime::lime(
  x = features,
  model = ensemble_tree, 
  n_bins = 10
)

class(components_lime)
## [1] "data_frame_explainer" "explainer"            "list"

summary(components_lime)
##                      Length Class              Mode     
## model                 1     H2ORegressionModel S4       
## preprocess            1     -none-             function 
## bin_continuous        1     -none-             logical  
## n_bins                1     -none-             numeric  
## quantile_bins         1     -none-             logical  
## use_density           1     -none-             logical  
## feature_type         80     -none-             character
## bin_cuts             80     -none-             list     
## feature_distribution 80     -none-             list
  1. Perform the LIME algorithm using the lime::explain() function on the observation(s) of interest.
# Use LIME to explain previously defined instances: high_ob and low_ob
lime_explanation <- lime::explain(
  # Observation(s) you want to create local explanations for
  x = rbind(high_ob, low_ob), 
  # Takes the explainer object created by lime::lime()
  explainer = components_lime, 
  # The number of permutations to create for each observation in x
  n_permutations = 5000,
  # The distance function to use `?dist()`
  dist_fun = "gower",
  # the distance measure to a similarity score
  kernel_width = 0.25,
  # The number of features to best describe
  # the predicted outcomes
  n_features = 10, 
  feature_select = "highest_weights"
)

glimpse(lime_explanation)
## Observations: 20
## Variables: 11
## $ model_type       <chr> "regression", "regression", "regression", "regr…
## $ case             <chr> "1825", "1825", "1825", "1825", "1825", "1825",…
## $ model_r2         <dbl> 0.41661172, 0.41661172, 0.41661172, 0.41661172,…
## $ model_intercept  <dbl> 186253.6, 186253.6, 186253.6, 186253.6, 186253.…
## $ model_prediction <dbl> 406033.5, 406033.5, 406033.5, 406033.5, 406033.…
## $ feature          <chr> "Gr_Liv_Area", "Overall_Qual", "Total_Bsmt_SF",…
## $ feature_value    <int> 3627, 8, 1930, 35760, 1796, 1831, 3, 14, 1, 3, …
## $ feature_weight   <dbl> 55254.859, 50069.347, 40261.324, 20430.128, 193…
## $ feature_desc     <chr> "2141 < Gr_Liv_Area", "Overall_Qual = Very_Exce…
## $ data             <list> [[Two_Story_1946_and_Newer, Residential_Low_De…
## $ prediction       <dbl> 663136.38, 663136.38, 663136.38, 663136.38, 663…

plot_features(lime_explanation, ncol = 1)

  1. Tune the LIME algorithm.
 # Tune the LIME algorithm a bit
lime_explanation2 <- explain(
  x = rbind(high_ob, low_ob), 
  explainer = components_lime, 
  n_permutations = 5000,
  dist_fun = "euclidean",
  kernel_width = 0.75,
  n_features = 10, 
  feature_select = "lasso_path"
)

# Plot the results
plot_features(lime_explanation2, ncol = 1)