CODEWITHSUNDEEP
pythonpandasscikit-learnscikits
MAK
MAK

Reputation: 615

MinMax scaler on list of dataframes

I have list of dataframes (all the dataframes has identical numeric columns ,represent different results of the same test. I want to keep them separated ). I want to training scikit MinMaxScalar that will take into account the minimum and maximum values for each column from all dataframes. May someone have solution to that?

Thanks,

MAK

Upvotes: 0

Views: 2782

Answers (2)

MaxU - stand with Ukraine
MaxU - stand with Ukraine

Reputation: 210832

You want to do the following:

  1. create a temporary DataFrame tmp as a concatenation of all your DFs from the list
  2. fit the MinMaxScaler object on tmp DF
  3. scale (transform) all DFs in the list using fitted MinMaxScaler object

UPDATE:

May you have a suggestion for training without creating temp dataframe?

we can make use of the .partial_fit() method in order to fit data from all DFs iteratively:

creating a list of sample DFs:

In [100]: dfs = [pd.DataFrame(np.random.rand(3,3)*100 - 50) for _ in range(3)]

In [101]: dfs[0]
Out[101]:
           0          1          2
0  45.473162  42.366712  41.395652
1 -35.476703  43.777850 -36.363200
2   0.479528  14.861075   4.196630

In [102]: dfs[2]
Out[102]:
           0          1          2
0   6.888876 -24.454986 -39.794309
1  -8.988094 -34.426252 -24.760782
2  34.317689 -43.644643  44.243769

scaling:

In [103]: from sklearn.preprocessing import MinMaxScaler

In [104]: mms = MinMaxScaler()

In [105]: _ = [mms.partial_fit(df) for df in dfs]

In [106]: scaled = [mms.transform(df) for df in dfs]

result:

In [107]: scaled[0]
Out[107]:
array([[1.        , 0.9838584 , 0.91065751],
       [0.07130264, 1.        , 0.03848462],
       [0.48381052, 0.66922958, 0.49341912]])

In [108]: scaled[1]
Out[108]:
array([[0.53340314, 0.8729412 , 0.62360548],
       [0.        , 0.39480025, 1.        ],
       [0.04767918, 0.10412712, 0.95859434]])

In [109]: scaled[2]
Out[109]:
array([[0.55734177, 0.2195048 , 0.        ],
       [0.37519322, 0.10544644, 0.16862177],
       [0.87201883, 0.        , 0.94260309]])

Upvotes: 2

Valdi_Bo
Valdi_Bo

Reputation: 30971

Look at the desription of methods of MinMaxScaler, e.g. in https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html

fit(X[, y])     Compute the minimum and maximum to be used for later scaling.
transform(X)    Scaling features of X according to feature_range.

So, assuming that your list of DataFrames is in dfList, probably you should:

  • Call fit(pd.concat(dfList)) to set min/max values, based on result of all experiments.
  • Then call transform(...) on each DataFrame.

Note that there is no explicit creation of any temporary table holding concatenated results of all experiments.

Upvotes: 0

Related Questions