CODEWITHSUNDEEP
pythonscikit-learn
user9799326
user9799326

Reputation:

sklearn MinMaxScaler: Inverse does not equal original

Consider the following python3 example where we use MinMaxScaler from scikit-learn to normalize a range of numbers, and then de-normalized them back to their original values.

from pandas import Series
from sklearn.preprocessing import MinMaxScaler

data = [10.0, 20.0, 30.0, 40.0, 50.0, 60.0, 70.0, 80.0, 90.0, 100.0]
series = Series(data)

values = series.values
values = values.reshape((len(values), 1))

scaler = MinMaxScaler(feature_range=(0,1))
scaler = scaler.fit(values)

normalized = scaler.transform(values)
inversed = scaler.inverse_transform(normalized)

One would expect that inversed equals values. Alas:

>>> inversed == values
array([[ True],
       [ True],
       [False],
       [ True],
       [ True],
       [False],
       [ True],
       [ True],
       [ True],
       [ True]])

>>> print(values)
[[ 10.]
 [ 20.]
 [ 30.]
 [ 40.]
 [ 50.]
 [ 60.]
 [ 70.]
 [ 80.]
 [ 90.]
 [100.]]

>>> print(inversed)
[[ 10.]
 [ 20.]
 [ 30.]
 [ 40.]
 [ 50.]
 [ 60.]
 [ 70.]
 [ 80.]
 [ 90.]
 [100.]]

What's happening here? Why is inversed[2] and inversed[5] unequal to values[2] and values[5]?

Upvotes: 2

Views: 2172

Answers (1)

Quickbeam2k1
Quickbeam2k1

Reputation: 5437

You are comparing two floats with equals. This might yield unexpected results due to rounding differences I would assume. Could you provide the bit patterns of the floating point values? You also might want to check the floating point guide.

As @CoMartel suggested, you can observe the difference between values and inverted (I haven't found a bit representation of floats in numpy, so far), you obtain

values - inversed
array([[ 0.00000000e+00],
       [ 0.00000000e+00],
       [-3.55271368e-15],
       [ 0.00000000e+00],
       [ 0.00000000e+00],
       [-7.10542736e-15],
       [ 0.00000000e+00],
       [ 0.00000000e+00],
       [ 0.00000000e+00],
       [ 0.00000000e+00]])

and you see that the corresponding values can not be the same

Upvotes: 1

Related Questions