I'm trying to use lime in order to explain text classification models in R (V4.0.5, Windows). With the logistic regression (library RWeka 0.4-43), I have no problems. The code
m_wlogistic <- Logistic(ID ~ ., data = training) explainer <- lime(sentence_to_explain, model = as_classifier(m_wlogistic), preprocess = testdata_preprocessor) explanation <- explain(sentence_to_explain, explainer, n_labels = 1, n_features = 2)
works. However, when I use a kernel SVM (library kernlab 0.9-29), I get a warning message for the lime command and an error for the explain command:
m_svm <- ksvm(ID ~ ., data = training, kernel = "rbfdot", C=rbf_C, kpar=list(sigma=rbf_sigma)) explainer <- lime(sentence_to_explain, model = as_classifier(m_svm), preprocess = testdata_preprocessor) explanation <- explain(sentence_to_explain, explainer, n_labels = 1, n_features = 2)
The warning message is:
Warning message: In class(x) <- c("lime_classifier", class(x)) : Setting class(x) to multiple strings ("lime_classifier", "ksvm", ...); result will no longer be an S4 object
The error is as follows:
Error in UseMethod("predict") : no applicable method for "predict" applied to an object of class "c('lime_classifier', 'ksvm')"
As I found out, there are methods to handle this with commands like "model_type()" and "predict_model()". However, unfortunately I didn't understand that approach and I didn't find a working example yet.
I have created a minimal runnable (respectively reproducing the error) example. Since it also needs pre-trained models and I cannot provide the training data, I have saved a test model as RDS and also created a text file as requested. The program, the data and the needed RDS files can be downloaded from the link below.