Examples

Featured Tutorials

PyThresh has a variety of different thresholding methods and an online example can be found at.

Towards Data Science: Thresholding Outlier Detection Scores with PyThresh

---

Karcher Mean Example

Full example: karch_example.py

1. Import models

   from pyod.models.knn import KNN
   from pyod.utils.data import generate_data
   
   from pyod.utils.data import evaluate_print
   from pyod.utils.example import visualize
   
   from pythresh.thresholds.karch import KARCH
   

2. Generate sample data with pyod.utils.data.generate_data():

   contamination = 0.1  # percentage of outliers
   n_train = 200  # number of training points
   n_test = 100  # number of testing points
   
   X_train, X_test, y_train, y_test = generate_data(
       n_train=n_train,
       n_test=n_test,
       n_features=2,
       contamination=contamination,
       random_state=42,
   )
   

3. Initialize a pyod.models.knn.KNN detector, fit the model, and threshold the outlier detection scores.

# train kNN detector
clf_name = "KNN"
clf = KNN()
clf.fit(X_train)
thres = KARCH()

# get the prediction labels and outlier scores of the training data
y_train_scores = clf.decision_scores_  # raw outlier scores
y_train_pred = thres.eval(y_train_scores)  # binary labels (0: inliers, 1: outliers)

# get the prediction on the test data
y_test_scores = clf.decision_function(X_test)  # outlier scores
y_test_pred = thres.eval(y_test_scores)  # outlier labels (0 or 1)

# it is possible to get the prediction confidence as well
y_test_pred, y_test_pred_confidence = clf.predict(
    X_test, return_confidence=True
)  # outlier labels (0 or 1) and confidence in the range of [0,1]

4. Evaluate the prediction using ROC and Precision @ Rank n pyod.utils.data.evaluate_print().

   from pyod.utils.data import evaluate_print
   
   # evaluate and print the results
   print("\nOn Training Data:")
   evaluate_print(clf_name, y_train, y_train_scores)
   print("\nOn Test Data:")
   evaluate_print(clf_name, y_test, y_test_scores)
   

5. See sample outputs on both training and test data.

   On Training Data:
   KNN ROC:0.9992, precision @ rank n:0.95
   
   On Test Data:
   KNN ROC:1.0, precision @ rank n:1.0
   

6. Generate the visualizations by visualize function included in all examples.

   visualize(
       clf_name,
       X_train,
       y_train,
       X_test,
       y_test,
       y_train_pred,
       y_test_pred,
       show_figure=True,
       save_figure=False,
   )
   

---

Model Combination Example

Just as outlier detection often suffers from model instability, a thresholding method may as well due to its unsupervised nature. Thus, it is recommended to combine various thresholders outputs, e.g., by averaging, to improve its robustness. Luckily this has already been written for convenience as the function pythresh.thresholds.comb.COMB

---

Additional API Example

1. Get the normalized threshold value that separates the inliers from outliers after the likelihood scores have been evaluated. Note, the outlier detection likelihood scores are normalized between 0 and 1.

# train kNN detector
clf_name = "KNN"
clf = KNN()
clf.fit(X_train)

scores = clf.decision_function(X_train)
thres = OCSVM()
labels = thres.eval(scores)

threshold = thres.thresh_

2. This can also be done for multiple outlier detector likelihood scores sets. These scores are first decomposed to 1D using a TruncatedSVD decomposition method. This decomposed score sets can also be accessed as a stores variable dscores_

   # train multiple detectors
   clf_name = "Multiple"
   clfs = [KNN(), IForest(), PCA()]
   
   scores = []
   for clf in clfs:
      clf.fit(X_train)
      scores.append(clf.decision_function(X_train))
   
   scores = np.vstack(scores).T
   
   thres = OCSVM()
   labels = thres.eval(scores)
   
   threshold = thres.thresh_
   dscores = thres.dscores_
   

3. Similarly, the lower and upper confidence interval of the contamination level for the pythresh.thresholds.comb.COMB thresholder can be retrieved.

# train kNN detector
clf_name = "KNN"
clf = KNN()
clf.fit(X_train)

scores = clf.decision_function(X_train)
thres = COMB()
labels = thres.eval(scores)

conf_interval = thres.confidence_interval_

For Jupyter Notebooks, please navigate to notebooks for additional use case references

References