monotonic2

Load the Yandex dataset with monotonic constraints. This dataset does not contain categorical features.

This dataset can be used for regression.

The contents of columns depends on the name or on the pattern of the name of the corresponding column:

  • Target(the first column) — Target values.

  • MonotonicNeg* — Monotonic negative numerical features.

    If values of such features decrease, then the prediction value must not decrease. Thus, if there are two objects x1x_{1} and x2x_{2} with all features being equal except for a monotonic negative feature MNegMNeg, such that x1[MNeg]>x2[MNeg]x_{1}[MNeg] > x_{2}[MNeg], then the following inequality must be met for predictions:

    f(x1)f(x2)f(x_{1}) \leq f(x_{2})

  • MonotonicPos* — Monotonic positive numerical features.

    If values of such features decrease, then the prediction value must not increase. Thus, if there are two objects x1x_{1} and x2x_{2} with all features being equal except for a monotonic positive feature MPosMPos, such that x1[MPos]>x2[MPos]x_{1}[MPos] > x_{2}[MPos], then the following inequality must be met for predictions:

    f(x1)f(x2)f(x_{1}) \geq f(x_{2})

Method call format

monotonic2()

Type of return value

A two pandas.DataFrame tuple (for train and validation datasets).

Usage examples

from catboost.datasets import monotonic2
monotonic2_train, monotonic2_test = monotonic2()

print(monotonic2_train.head(3))

The output of this example:

   Target  MonotonicNeg0  MonotonicPos0  MonotonicPos1  MonotonicNeg1
0     0.0            NaN            NaN       0.010356       0.032638
1     0.0            NaN            NaN       0.010356       0.032638
Previous
Next