Blogpost-Changepoint-Detection-Snippet4

ChangepointDetection4.py

from sklearn.base import BaseEstimator, TransformerMixin
from sklearn.pipeline import Pipeline, make_pipeline, FeatureUnion, make_union
from sklearn.ensemble import RandomForestRegressor

def get_categorical_glossary(x):
    n_unique = len(np.unique(x[x.notnull()]))
    codes = range(0, n_unique)
    return dict(zip(np.unique(x), codes))

class ColumnSelector(BaseEstimator, TransformerMixin):
    """
    Selects columns by name
    """    
    def __init__(self, colnames):
        self.colnames = colnames

    def fit(self, X, y=None):
        return self

    def transform(self, X, y=None):
        return X[self.colnames]
    
class NumericEncoder(BaseEstimator, TransformerMixin):
    """
    Replaces categoricals with integer indices.
    We need this since the default scikit-learn LabelEncoder is not supported in Pipelines.
    """
    def fit(self, X, y=None):
        self.glossary = dict([(nm, get_categorical_glossary(X[nm])) for nm in X])
        return self

    def transform(self, X, y=None):
        X_new = X.copy()
        for nm in X:
            # if missing in dict, sets to NaN
            X_new[nm] = X[nm].map(self.glossary[nm])
        return X_new