State which three libraries the pipeline imports, their conventional aliases, and what each is used for downstream.

• numpy as np — numerical arrays and math; used later for np.nan (the missing-value marker) and np.array(...).
• matplotlib.pyplot as plt — plotting library, imported by convention as part of the standard template even though this script never plots.
• pandas as pd — tabular data loading and manipulation; provides pd.read_csv() and the DataFrame.

Explain how the dataset is read and how the feature matrix X and dependent vector y are created.

• pd.read_csv('Data.csv') — reads the file into a pandas DataFrame stored in dataset.
• X = dataset.iloc[:, :-1].values — every row, every column except the last → the feature matrix (Country, Age, Salary).
• y = dataset.iloc[:, -1].values — every row, only the last column → the dependent variable (Purchased).
• .iloc / .values — .iloc selects by integer position (not label); .values returns the underlying NumPy array that scikit-learn expects.

Describe how missing Age/Salary values are replaced and why only those columns are touched.

• from sklearn.impute import SimpleImputer — the imputer class lives in sklearn.impute.
• SimpleImputer(missing_values=np.nan, strategy='mean') — replaces each NaN with the mean of its column.
• X[:, 1:3] — applied only to columns 1–2 (Age, Salary); column 0 (Country) is text and has no mean.

State the difference between fit, transform, and fit_transform and why order matters.

• fit() — learns parameters from the data (here, the column means) but changes nothing.
• transform() — applies those learned parameters to actually replace the values.
• fit_transform() — does both in one call; the imputer example separates them (imputer.fit(...) then imputer.transform(...)) to make the two steps explicit.

Explain how the Country column is one-hot encoded and what each argument does.

• ColumnTransformer (from sklearn.compose) — applies a transformer to specific columns only.
• transformers=[('encoder', OneHotEncoder(), [0])] — one-hot encode column [0] (Country); OneHotEncoder is from sklearn.preprocessing.
• remainder='passthrough' — keep all other columns (Age, Salary) unchanged instead of dropping them.
• X = np.array(ct.fit_transform(X)) — run it and force the result back into a NumPy array.

Say why one-hot encoding is used instead of numeric labels, and where the resulting columns appear.

• Why one-hot — labeling France/Germany/Spain as 0/1/2 would imply a false order/magnitude; one-hot avoids that by making each category its own 0/1 column.
• Three new columns — France, Germany, Spain become three binary dummy columns.
• Position — the dummy columns are placed at the front (columns 0–2); Age and Salary shift to become columns 3 and 4.

Explain how the Purchased column is encoded and why LabelEncoder is appropriate here.

• LabelEncoder (from sklearn.preprocessing) — converts category labels to integers.
• y = le.fit_transform(y) — turns No/Yes into 0/1 (alphabetical: No=0, Yes=1).
• Why it's fine for y — the dependent variable is a single target column, so a false ordinal order is not a concern the way it is for feature columns.

Describe train_test_split and what its key arguments control.

• from sklearn.model_selection import train_test_split — the split function's module.
• test_size = 0.2 — 20% of rows go to the test set, 80% to training.
• Four outputs — X_train, X_test, y_train, y_test in that fixed order.

Say what random_state does and why it is fixed.

• What it does — train_test_split shuffles the rows before splitting (shuffle=True by default); random_state seeds that shuffle so the same split happens every run.
• Any integer works — the specific value (0, 1, 42) is arbitrary; what matters is using the same one consistently.
• This module — the worked file uses random_state = 1; the template uses 0. Different seeds give different rows in each set.

Explain what standardization does and how StandardScaler is applied.

• StandardScaler (from sklearn.preprocessing) — standardizes values to mean 0, unit variance: z = (x − mean) / std.
• Why — puts Age and Salary on the same scale so large-magnitude features (Salary) don't dominate models that are distance/gradient based.
• Applied after the split — scaling comes last, on the already-split data.

Explain the two subtle choices in the scaling step.

• Only columns [:, 3:] — scaling is applied to Age/Salary (columns 3+), not the one-hot dummy columns, which are already 0/1 and should stay interpretable.
• fit_transform on X_train, transform on X_test — the scaler learns mean/std from the training set only, then re-uses them on the test set.
• Why — fitting on the test set too would leak test information into training (data leakage) and give an over-optimistic evaluation.