Using numerical and categorical variables together

In the previous notebooks, we showed the required preprocessing to apply when dealing with numerical and categorical variables. However, we decoupled the process to treat each type individually. In this notebook, we show how to combine these preprocessing steps.

We first load the entire adult census dataset.

import pandas as pd

adult_census = pd.read_csv("../datasets/adult-census.csv")
# drop the duplicated column `"education-num"` as stated in the first notebook
adult_census = adult_census.drop(columns="education-num")

target_name = "class"
target = adult_census[target_name]

data = adult_census.drop(columns=[target_name])

Selection based on data types

We separate categorical and numerical variables using their data types to identify them, as we saw previously that object corresponds to categorical columns (strings). We make use of make_column_selector helper to select the corresponding columns.

from sklearn.compose import make_column_selector as selector

numerical_columns_selector = selector(dtype_exclude=object)
categorical_columns_selector = selector(dtype_include=object)

numerical_columns = numerical_columns_selector(data)
categorical_columns = categorical_columns_selector(data)

Caution

Here, we know that object data type is used to represent strings and thus categorical features. Be aware that this is not always the case. Sometimes object data type could contain other types of information, such as dates that were not properly formatted (strings) and yet relate to a quantity of elapsed time.

In a more general scenario you should manually introspect the content of your dataframe not to wrongly use make_column_selector.

Dispatch columns to a specific processor

In the previous sections, we saw that we need to treat data differently depending on their nature (i.e. numerical or categorical).

Scikit-learn provides a ColumnTransformer class which sends specific columns to a specific transformer, making it easy to fit a single predictive model on a dataset that combines both kinds of variables together (heterogeneously typed tabular data).

We first define the columns depending on their data type:

• one-hot encoding is applied to categorical columns. Besides, we use handle_unknown="ignore" to solve the potential issues due to rare categories.

• numerical scaling numerical features which will be standardized.

Now, we create our ColumnTransfomer by specifying three values: the preprocessor name, the transformer, and the columns. First, let’s create the preprocessors for the numerical and categorical parts.

from sklearn.preprocessing import OneHotEncoder, StandardScaler

categorical_preprocessor = OneHotEncoder(handle_unknown="ignore")
numerical_preprocessor = StandardScaler()

Now, we create the transformer and associate each of these preprocessors with their respective columns.

from sklearn.compose import ColumnTransformer

preprocessor = ColumnTransformer(
    [
        ("one-hot-encoder", categorical_preprocessor, categorical_columns),
        ("standard_scaler", numerical_preprocessor, numerical_columns),
    ]
)

We can take a minute to represent graphically the structure of a ColumnTransformer:

A ColumnTransformer does the following:

• It splits the columns of the original dataset based on the column names or indices provided. We obtain as many subsets as the number of transformers passed into the ColumnTransformer.

• It transforms each subsets. A specific transformer is applied to each subset: it internally calls fit_transform or transform. The output of this step is a set of transformed datasets.

• It then concatenates the transformed datasets into a single dataset.

The important thing is that ColumnTransformer is like any other scikit-learn transformer. In particular it can be combined with a classifier in a Pipeline:

from sklearn.linear_model import LogisticRegression
from sklearn.pipeline import make_pipeline

model = make_pipeline(preprocessor, LogisticRegression(max_iter=500))
model

Pipeline(steps=[('columntransformer',
                 ColumnTransformer(transformers=[('one-hot-encoder',
                                                  OneHotEncoder(handle_unknown='ignore'),
                                                  ['workclass', 'education',
                                                   'marital-status',
                                                   'occupation', 'relationship',
                                                   'race', 'sex',
                                                   'native-country']),
                                                 ('standard_scaler',
                                                  StandardScaler(),
                                                  ['age', 'capital-gain',
                                                   'capital-loss',
                                                   'hours-per-week'])])),
                ('logisticregression', LogisticRegression(max_iter=500))])

In a Jupyter environment, please rerun this cell to show the HTML representation or trust the notebook.
On GitHub, the HTML representation is unable to render, please try loading this page with nbviewer.org.

The final model is more complex than the previous models but still follows the same API (the same set of methods that can be called by the user):

• the fit method is called to preprocess the data and then train the classifier of the preprocessed data;

• the predict method makes predictions on new data;

• the score method is used to predict on the test data and compare the predictions to the expected test labels to compute the accuracy.

Let’s start by splitting our data into train and test sets.

from sklearn.model_selection import train_test_split

data_train, data_test, target_train, target_test = train_test_split(
    data, target, random_state=42
)

Caution

Be aware that we use train_test_split here for didactic purposes, to show the scikit-learn API. In a real setting one might prefer to use cross-validation to also be able to evaluate the uncertainty of our estimation of the generalization performance of a model, as previously demonstrated.

Now, we can train the model on the train set.

_ = model.fit(data_train, target_train)

Then, we can send the raw dataset straight to the pipeline. Indeed, we do not need to make any manual preprocessing (calling the transform or fit_transform methods) as it is already handled when calling the predict method. As an example, we predict on the five first samples from the test set.

data_test.head()

	age	workclass	education	marital-status	occupation	relationship	race	sex	capital-gain	capital-loss	hours-per-week	native-country
7762	56	Private	HS-grad	Divorced	Other-service	Unmarried	White	Female	0	0	40	United-States
23881	25	Private	HS-grad	Married-civ-spouse	Transport-moving	Own-child	Other	Male	0	0	40	United-States
30507	43	Private	Bachelors	Divorced	Prof-specialty	Not-in-family	White	Female	14344	0	40	United-States
28911	32	Private	HS-grad	Married-civ-spouse	Transport-moving	Husband	White	Male	0	0	40	United-States
19484	39	Private	Bachelors	Married-civ-spouse	Sales	Wife	White	Female	0	0	30	United-States

model.predict(data_test)[:5]

array([' <=50K', ' <=50K', ' >50K', ' <=50K', ' >50K'], dtype=object)

target_test[:5]

7762      <=50K
23881     <=50K
30507      >50K
28911     <=50K
19484     <=50K
Name: class, dtype: object

To get directly the accuracy score, we need to call the score method. Let’s compute the accuracy score on the entire test set.

model.score(data_test, target_test)

0.8575055278028008

Evaluation of the model with cross-validation

As previously stated, a predictive model should be evaluated by cross-validation. Our model is usable with the cross-validation tools of scikit-learn as any other predictors:

from sklearn.model_selection import cross_validate

cv_results = cross_validate(model, data, target, cv=5)
cv_results

{'fit_time': array([0.24602199, 0.24946833, 0.21629882, 0.23841834, 0.25856304]),
 'score_time': array([0.02780533, 0.0276289 , 0.02771664, 0.02946305, 0.02800608]),
 'test_score': array([0.85116184, 0.84993346, 0.8482801 , 0.85257985, 0.85544636])}

scores = cv_results["test_score"]
print(
    "The mean cross-validation accuracy is: "
    f"{scores.mean():.3f} ± {scores.std():.3f}"
)

The mean cross-validation accuracy is: 0.851 ± 0.002

The compound model has a higher predictive accuracy than the two models that used numerical and categorical variables in isolation.

In this notebook we:

• used a ColumnTransformer to apply different preprocessing for categorical and numerical variables;

• used a pipeline to chain the ColumnTransformer preprocessing and logistic regression fitting.