python - 整个数据帧python的最小最大缩放

Question

我正在使用 from sklearn.preprocessing import MinMaxScaler
使用以下代码和数据集:

df = pd.DataFrame({
  "A" : [-0.5624105,
-0.5637749,
0.2973856,
0.619784,
0.007297921,
0.8146919,
0.1082434,
-0.2311236,
-0.6945567,
-0.6807524,
-0.1017431,
0.5889628,
0.5384794,
0.3906553,
0.3843442,
0.4408366,
0.4035791,
0.05258237,
-0.4847771
],
  "B" : [-0.5068743,
0.1422121,
0.6444226,
0.4959088,
-0.2260773,
0.3420533,
0.2346546,
0.1177824,
-0.7701161,
-0.7566853,
-0.5091485,
0.4509938,
0.4209853,
0.304058,
0.3753832,
0.6958977,
0.6763205,
0.05536954,
-0.9857719
]})

min_max_scaler = MinMaxScaler(feature_range=(0,255))

print(df)

df[df.columns] = min_max_scaler.fit_transform(df[df.columns])

print(df)
print(type(df))

我想用整个数据集中的最小值和整个数据集中的最大值来缩放它我如何使用相同的代码管理它？是否可以？

           A         B
0  -0.562411 -0.506874
1  -0.563775  0.142212
2   0.297386  0.644423
3   0.619784  0.495909
4   0.007298 -0.226077
5   0.814692  0.342053
6   0.108243  0.234655
7  -0.231124  0.117782
8  -0.694557 -0.770116
9  -0.680752 -0.756685
10 -0.101743 -0.509149
11  0.588963  0.450994
12  0.538479  0.420985
13  0.390655  0.304058
14  0.384344  0.375383
15  0.440837  0.695898
16  0.403579  0.676320
17  0.052582  0.055370
18 -0.484777 -0.985772
             A           B
0    22.327190   72.617646
1    22.096664  171.041874
2   167.596834  247.194572
3   222.068703  224.674680
4   118.584127  115.196304
5   255.000000  201.344798
6   135.639699  185.059394
7    78.300845  167.337476
8     0.000000   32.700971
9     2.332350   34.737551
10  100.160748   72.272798
11  216.861207  217.863993
12  208.331620  213.313653
13  183.355519  195.583380
14  182.289206  206.398778
15  191.834063  255.000000
16  185.539101  252.031411
17  126.235309  157.873501
18   35.443994    0.000000

我不希望每一列都有不同的映射，我需要使用 -0.985772 0.814692(b 列第 18 行，a 列第 5 行)来映射它

Answer 1

您有两种方法可以做到这一点:

# Manually:
min_value, max_value = df.min().min(), df.max().max()
scaled1 = (df - min_value) * 255 / (max_value - min_value)

# Using MinMaxScaler
min_max_scaler = MinMaxScaler(feature_range=(0,255))    
# Stack everything into a single column to scale by the global min / max
tmp = df.to_numpy().reshape(-1,1)
scaled2 = min_max_scaler.fit_transform(tmp).reshape(len(df), 2)

两者都返回相同的结果:

np.isclose(scaled1, scaled2).all()
# True

您可以使用缩放值创建一个新的 DataFrame:

scaled = pd.DataFrame(scaled1, index=df.index, columns=df.columns)

或者将它们分配回 df :

df.loc[:] = scaled1