MinMaxScaler only Generating Positive Values

Question

from sklearn import preprocessing

I have a data which contain positive and negative values as given below :

(Here is the csv file of the data https://drive.google.com/file/d/1-Kc3vGDazAPQ_4I7wVvG6VI9Bd9b4uCW/view?usp=sharing)

ext is: | Index | Values | | -------- | -------------- | | 1 | -5.473753 | | 2 | 54.730399 | | 3 | 0.389353 | | 4 | -4.156109 | | 5 | 65.108997 | | ... | ......... | | 733 | 14.082214 | | 734 | 107.248120 | | 735 | 54.730399 |

I am trying to use MinMaxScaler as given below:

min_max_scaler = preprocessing.MinMaxScaler()
test_scaled = min_max_scaler.fit_transform(ext)
predictions_rescaled=min_max_scaler.inverse_transform(test_scaled)

predictions_rescaled should be same as ext, because i am scaling it and then rescaling it, but surprisingly both are different. Can anyone guide me where i am making mistake in scaling-rescaling process.

Could you specify in you post framework used, I mean add required imports. Your goal is to provide a [mcve]. You are close to it. — jlandercy, Jan 24 '21 at 07:09
The problem is somewhere else, as I have used your data and could reverse it. See my code below. — Prayson W. Daniel, Jan 24 '21 at 07:46
I have updated my answer with suggestions to debug the issue. — Prayson W. Daniel, Jan 24 '21 at 07:54
For negative value - you probably to keep it negative. So split the scalare for negative values (max is zero) and positive values (min is zero). — Eitan, Nov 05 '22 at 21:02

Prakash Dahal · Accepted Answer · 2021-01-24T07:25:39.130

2

MinMaxScaler scales the values in range 0 to 1 by default. If you want negative numbers after scaling, you can use StandardScaler.

Also there is no wrong in your code. Inverse_transform() is returning the old dataframe.

s_scaler = MinMaxScaler()
test_scaled = s_scaler.fit_transform(ext)
print(test_scaled)
predictions_rescaled=s_scaler.inverse_transform(test_scaled)
print()
predictions_rescaled = pd.DataFrame(predictions_rescaled)
predictions_rescaled

edited Jan 24 '21 at 07:25

answered Jan 24 '21 at 07:16

Prakash Dahal

4,388
2
11
25

1

I dont want negative numbers on scaling, but after rescaling i must get original numbers (which do have some negative values as well) – Mohd Naved Jan 24 '21 at 07:19
No I am not getting original values back. Plz check. – Mohd Naved Jan 24 '21 at 07:22
Your current code is for standard scaler and not for MinMaxScaler. – Mohd Naved Jan 24 '21 at 07:25
I have updated it for MinMaxScaler. Is it fine now? – Prakash Dahal Jan 24 '21 at 07:26

Prayson W. Daniel · Answer 2 · 2021-01-24T07:53:37.933

1

Check which scikit-version you are using and whether there exists a bug in MinMaxScaler. If that is not the problem, check the way you pass in data to MinMaxScaler.

From scikit-learn documentation here, it should behaviour as expected.

I cannot reproduce your problem. It works fine. I get originally back:


import pandas as pd
from sklearn.preprocessing import MinMaxScaler

df = pd.read_csv('ext.csv', index_col=0)
scaler = MinMaxScaler()

df['minmax'] = scaler.fit_transform(df)
df['inv'] = scaler.inverse_transform(df[['minmax']])

enter image description here

edited Jan 24 '21 at 07:53

answered Jan 24 '21 at 07:15

Prayson W. Daniel

14,191
4
51
57

On scaling i am ok with positive numbers (in between 0 to 1), but when i rescaled these numbers i should be getting my original ext data (which contain positive and negative numbers), but this is not happening with the my current code. – Mohd Naved Jan 24 '21 at 07:20
I get the value back with sckit-learn v 0.23 – Prayson W. Daniel Jan 24 '21 at 07:41

MinMaxScaler only Generating Positive Values

2 Answers2