Source code for libpyhat.utils.outlier_identify

import numpy as np
from sklearn.ensemble import IsolationForest
from sklearn.neighbors import LocalOutlierFactor


# This function flags outliers in a spectral data set. Different algorithms
# can be selected.




def outlier_identify(df, col, method, params, meta_label="meta"):
    assert method in ["Isolation Forest", "Local Outlier Factor"]

    if method == "Isolation Forest":
        # Build object
        obj = IsolationForest(**params)

        # Perform fit
        fit = obj.fit(np.array(df[col]))

        # Get outlier scores
        outlier_scores = fit.decision_function(np.array(df[col]))

        # Save outlier scores to dataset
        df[(meta_label, "Outlier Scores - " + method + str(params))] = outlier_scores

        # Get outlier predictions
        is_outlier = obj.predict(np.array(df[col]))

        # Save predictions
        df[(meta_label, "Outliers - " + method + str(params))] = is_outlier

    elif method == "Local Outlier Factor":
        # Build object
        obj = LocalOutlierFactor(**params)

        # Perform fit
        fit = obj.fit(np.array(df[col]))

        # Get predictions
        is_outlier = obj.fit_predict(np.array(df[col]))

        # Save predictions and outlier factors
        df[
            (meta_label, "Outlier Factor - " + method + str(params))
        ] = obj.negative_outlier_factor_
        df[(meta_label, "Outliers - " + method + str(params))] = is_outlier
    return df