CODEWITHSUNDEEP
pythonscikit-learnjoblib
Bobs Burgers
Bobs Burgers

Reputation: 821

AttributeError: 'MinMaxScaler' object has no attribute 'clip'

I get the following error when I attempt to load a saved sklearn.preprocessing.MinMaxScaler

/shared/env/lib/python3.6/site-packages/sklearn/base.py:315: UserWarning: Trying to unpickle estimator MinMaxScaler from version 0.23.2 when using version 0.24.0. This might lead to breaking code or invalid results. Use at your own risk.
  UserWarning)
[2021-01-08 19:40:28,805 INFO train.py:1317 - main ] EXCEPTION WORKER 100: 
Traceback (most recent call last):
  ...
  File "/shared/core/simulate.py", line 129, in process_obs
    obs = scaler.transform(obs)
  File "/shared/env/lib/python3.6/site-packages/sklearn/preprocessing/_data.py", line 439, in transform
    if self.clip:
AttributeError: 'MinMaxScaler' object has no attribute 'clip'

I trained the scaler on one machine, saved it, and pushed it to a second machine where it was loaded and used to transform input.

# loading and transforming
import joblib
from sklearn.preprocessing import MinMaxScaler

scaler = joblib.load('scaler')
assert isinstance(scaler, MinMaxScaler)
data = scaler.transform(data)  # throws exception

Upvotes: 8

Views: 17004

Answers (5)

Marcelo Soares
Marcelo Soares

Reputation: 51

An approach that worked for me was to monkey-patch the MinMaxScaler.__setstate__ to set this attribute manually to false if non-existent with the snippet below:

from typing import Any, Dict

from sklearn.preprocessing import MinMaxScaler


original_minmax_setstate = MinMaxScaler.__setstate__

def __monkey_patch_minmax_setstate__(self, state: Dict[str, Any]) -> None:
    state.setdefault("clip", False)
    original_minmax_setstate(self, state)

MinMaxScaler.__setstate__ = __monkey_patch_minmax_setstate__


# Deserialize the scaler here

By adding this above the deserialization (unpickling), I managed to deserialize my object in a newer scikit-learn version and then re-serialize it with the clip alteration, no longer needing the monkey-patch afterwards.

Upvotes: 1

Muhammad Fayzan
Muhammad Fayzan

Reputation: 91

version issue of sklearn
You need to install in windows
pip install scikit-learn==0.24.0
I solve my Problem using this command

Upvotes: 0

Peter Trcka
Peter Trcka

Reputation: 1531

New property clip was added to MinMaxScaler in later version (since 0.24).

# loading and transforming
import joblib
from sklearn.preprocessing import MinMaxScaler

scaler = joblib.load('scaler')
assert isinstance(scaler, MinMaxScaler)
scaler.clip = False  # add this line
data = scaler.transform(data)  # throws exceptio

Explanation:

Becase clip is defined in __init__ method it is part of MinMaxScaler.__dict__. When you try to create object from pickle __setattr__ method is used to set all attributues, but clip was not used in older version therefore is missing in your new MinMaxScale instance. Simply add:

scaler.clip = False

and it should work fine.

Upvotes: 5

Sau
Sau

Reputation: 119

I solved this issue with pip install scikit-learn==0.23.2 in my conda or cmd. Essentially downgrading the scikit module helped.

Upvotes: 4

Bobs Burgers
Bobs Burgers

Reputation: 821

The issue is you are training the scaler on a machine with an older verion of sklearn than the machine you're using to load the scaler.

Noitce the UserWarning

UserWarning: Trying to unpickle estimator MinMaxScaler from version 0.23.2 when using version 0.24.0. This might lead to breaking code or invalid results. Use at your own risk. UserWarning)

The solution is to fix the version mismatch. Either by upgrading one sklearn to 0.24.0 or downgrading to 0.23.2

Upvotes: 12

Related Questions