Feature Engineering Tasks#

We’ll define the relevant feature engineering tasks to clean up the SQLite data.

First, let’s import the required libraries.

import numpy as np
import pandas as pd
from flytekit import task
from flytekit.types.schema import FlyteSchema
from numpy.core.fromnumeric import sort
from sklearn.feature_selection import SelectKBest, f_classif
from sklearn.impute import SimpleImputer

There are a specific set of columns for which imputation isn’t required. We ignore them.

    "Hospital Number",
    "surgical lesion",

We define a mean_median_imputer task to fill in the missing values of the dataset, for which we use the SimpleImputer class from the scikit-learn library.

@task(cache=True, cache_version="1.0")
def mean_median_imputer(
    dataframe: pd.DataFrame,
    imputation_method: str,
) -> FlyteSchema:
    dataframe = dataframe.replace("?", np.nan)
    if imputation_method not in ["median", "mean"]:
        raise ValueError("imputation_method takes only values 'median' or 'mean'")

    imputer = SimpleImputer(missing_values=np.nan, strategy=imputation_method)

    imputer = imputer.fit(
    ] = imputer.transform(
    return dataframe

Let’s define the other task called univariate_selection that does feature selection. The SelectKBest method removes all but the highest scoring features (DataFrame columns).

@task(cache=True, cache_version="1.0")
def univariate_selection(
    dataframe: pd.DataFrame, num_features: int, data_class: str
) -> pd.DataFrame:
    # Remove ``timestamp`` and ``Hospital Number`` columns as they ought to be present in the dataset
    dataframe = dataframe.drop(["event_timestamp", "Hospital Number"], axis=1)

    if num_features > 9:
        raise ValueError(
            f"Number of features must be <= 9; you've given {num_features}"

    X = dataframe.iloc[:, dataframe.columns != data_class]
    y = dataframe.loc[:, data_class]
    test = SelectKBest(score_func=f_classif, k=num_features)
    fit = test.fit(X, y)
    indices = sort((-fit.scores_).argsort()[:num_features])
    column_names = list(map(X.columns.__getitem__, indices))
    features = fit.transform(X)
    return pd.DataFrame(np.c_[features, y.to_numpy()], columns=column_names)

Total running time of the script: ( 0 minutes 0.000 seconds)

Gallery generated by Sphinx-Gallery