Minimal example of RDST with visualisation

This example give the minimal setup to run RDST and visualize the result

on a dataset from the UCR archive.

Load the dataset and run RDST with Ridge

We load a UCR dataset with its name, and initialise RDST with a Ridge classifier using the wrapper class R_DST_Ridge.

from convst.classifiers import R_DST_Ridge
from convst.utils.dataset_utils import load_sktime_dataset_split

X_train, X_test, y_train, y_test, _ = load_sktime_dataset_split('GunPoint')

rdst = R_DST_Ridge(n_shapelets=10000, n_jobs=1)

rdst.fit(X_train, y_train)
acc_score = rdst.score(X_test, y_test)

print("Accuracy Score for RDST : {}".format(acc_score))

Visualize the best shapelet of a class

To visualize a shapelet, we use the dedicated interpreter class. It uses the coefficients from the Ridge classifier to select a shapelet that was important for the classification task for one class. We can then plot the distribution of features the shapelet generate for both the training and the testing set, too see if the distributions are alike.

from convst.interpreters import RDST_Ridge_interpreter
target_class = 1
interpreter = RDST_Ridge_interpreter(rdst)

interpreter.visualize_best_shapelets_one_class(
    X_train, y_train, target_class, n_shp=1
)

interpreter.visualize_best_shapelets_one_class(
    X_test, y_test, target_class, n_shp=1
)