Transformer pipeline

The problem, from first principles

Standard scaler without leakage showed why preprocessing must learn its statistics from training data only. Real projects chain several preprocessing steps, and doing that by hand invites mistakes. A pipeline bundles the steps in order: fitting the pipeline fits each step on the output of the previous one, and transforming new data replays the same sequence with the fitted numbers.

Implement two classes.

• MinMaxScaler: fit(X) stores each column's minimum and maximum. transform(X) returns (X - min) / (max - min). A column whose training range is zero maps every value to 0. fit_transform(X) fits, then transforms X, and returns the result.
• Pipeline(steps): steps is a list of transformers. fit_transform(X) runs each step's fit_transform in order and returns the final output. transform(X) runs each step's transform in order.

Constraints

• transform never refits anything. Values outside the training range may fall outside [0, 1], and that is correct.
• fit returns the scaler itself, so calls can be chained.
• Do not modify the inputs.

Hints

Open one at a time. Each gives away a little more than the last.