Handling missing data

Supervised Learning with scikit-learn

George Boorman

Core Curriculum Manager, DataCamp

Missing data

  • No value for a feature in a particular row

  • This can occur because:

    • There may have been no observation
    • The data might be corrupt

  • We need to deal with missing data

Supervised Learning with scikit-learn

Music dataset

print(music_df.isna().sum().sort_values())

genre                 8
popularity           31
loudness             44
liveness             46
tempo                46
speechiness          59
duration_ms          91
instrumentalness     91
danceability        143
valence             143
acousticness        200
energy              200
dtype: int64
Supervised Learning with scikit-learn

Dropping missing data

music_df = music_df.dropna(subset=["genre", "popularity", "loudness", "liveness", "tempo"])

print(music_df.isna().sum().sort_values())

popularity            0
liveness              0
loudness              0
tempo                 0
genre                 0
duration_ms          29
instrumentalness     29
speechiness          53
danceability        127
valence             127
acousticness        178
energy              178
dtype: int64
Supervised Learning with scikit-learn

Imputing values

  • Imputation - use subject-matter expertise to replace missing data with educated guesses
  • Common to use the mean
  • Can also use the median, or another value
  • For categorical values, we typically use the most frequent value - the mode
  • Must split our data first, to avoid data leakage
Supervised Learning with scikit-learn

Imputation with scikit-learn

from sklearn.impute import SimpleImputer

X_cat = music_df["genre"].values.reshape(-1, 1)
X_num = music_df.drop(["genre", "popularity"], axis=1).values
y = music_df["popularity"].values

X_train_cat, X_test_cat, y_train, y_test = train_test_split(X_cat, y, test_size=0.2,
                                                            random_state=12)

X_train_num, X_test_num, y_train, y_test = train_test_split(X_num, y, test_size=0.2,
                                                            random_state=12)

imp_cat = SimpleImputer(strategy="most_frequent")

X_train_cat = imp_cat.fit_transform(X_train_cat)

X_test_cat = imp_cat.transform(X_test_cat)
Supervised Learning with scikit-learn

Imputation with scikit-learn

imp_num = SimpleImputer()

X_train_num = imp_num.fit_transform(X_train_num)

X_test_num = imp_num.transform(X_test_num)

X_train = np.append(X_train_num, X_train_cat, axis=1)

X_test = np.append(X_test_num, X_test_cat, axis=1)
  • Imputers are known as transformers
Supervised Learning with scikit-learn

Imputing within a pipeline

from sklearn.pipeline import Pipeline

music_df = music_df.dropna(subset=["genre", "popularity", "loudness", "liveness", "tempo"])

music_df["genre"] = np.where(music_df["genre"] == "Rock", 1, 0)

X = music_df.drop("genre", axis=1).values
y = music_df["genre"].values
Supervised Learning with scikit-learn

Imputing within a pipeline

steps = [("imputation", SimpleImputer()),
         ("logistic_regression", LogisticRegression())]

pipeline = Pipeline(steps)

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3,
                                                    random_state=42)

pipeline.fit(X_train, y_train)

pipeline.score(X_test, y_test)

0.7593582887700535
Supervised Learning with scikit-learn

Let's practice!

Supervised Learning with scikit-learn

Preparing Video For Download...