MinMaxScaler

MinMaxScaler compresses data between 0 and 1.

MinMaxScaler

MinMaxScaler moves each feature into the [0, 1] range with (x - min) / (max - min). The smallest value becomes 0 and the largest 1:

from sklearn.preprocessing import MinMaxScaler
mm = MinMaxScaler()
print(mm.fit_transform([[10], [20], [30]]))   # [[0. ] [0.5] [1. ]]

It is handy where a bounded range is expected, like image pixels (0-255) and neural network inputs. The range can be changed with feature_range=(-1, 1).

Outlier sensitivity and RobustScaler

Since MinMax relies on min and max, a single outlier squeezes all the other values into a narrow range. RobustScaler uses the median and the interquartile range:

x = np.array([[1], [2], [3], [4], [100]])
print(MinMaxScaler().fit_transform(x).ravel().round(2))   # [0.   0.01 0.02 0.03 1.  ]
from sklearn.preprocessing import RobustScaler
print(RobustScaler().fit_transform(x).ravel())            # [-1.  -0.5  0.   0.5 48.5]

With RobustScaler, typical values stay in a sensible range and the outlier stands out as large.

Exercise Two scalers

Transform X with MinMaxScaler and print it. Then transform the example [[30, 7500]] with the same scaler and print it.

Show example solution

import numpy as np
from sklearn.preprocessing import StandardScaler, MinMaxScaler

X = np.array([[25, 3000], [35, 5000], [45, 10000]], dtype=float)

mm = MinMaxScaler()
print(mm.fit_transform(X))
print(mm.transform([[30, 7500]]))

Test yourself

For a feature with min=0 and max=200, what does 50 become with MinMaxScaler?
- 0.2
- 0.25
- 0.5
- 50
Show answer

0.25 — (50 - 0) / (200 - 0) = 0.25.

Move the values into the range -1 to 1.

mm = MinMaxScaler(____=(-1, 1))

Show answer

mm = MinMaxScaler(feature_range=(-1, 1))

Match the scalers.

StandardScaler

Mean 0, std 1
MinMaxScaler

[0, 1] range
RobustScaler

Median and IQR
feature_range

Target range

Next lesson: When to use?

We Are Enhancing Your Experience

Cookie Preferences

Strictly Necessary Cookies

Analytical Cookies