如何在scikit学习回归中不标准化目标数据

Question

I am trying to predict future profit data in a dataset of a copper mine enterprise data in csv format.

I read the data:

data = pd.read_csv('data.csv')

I split the data:

data_target = data[target].astype(float)
data_used = data.drop(['Periodo', 'utilidad_operativa_dolar'], axis=1)
x_train, x_test, y_train, y_test = train_test_split(data_used, data_target, test_size=0.4,random_state=33)

Create an svr predictor:

clf_svr= svm.SVR(kernel='rbf')

Standarize the data:

from sklearn.preprocessing import StandardScaler
scalerX = StandardScaler().fit(x_train)
scalery = StandardScaler().fit(y_train)

x_train = scalerX.transform(x_train)
y_train = scalery.transform(y_train)
x_test = scalerX.transform(x_test)
y_test = scalery.transform(y_test)

print np.max(x_train), np.min(x_train), np.mean(x_train), np.max(y_train), np.min(y_train), np.mean(y_train)

Then predict:

y_pred=clf.predict(x_test)

And the prediction data is standarized as well. I want the predicted data to be in the original format, how i can do that?

Answer 1

You would want to use the inverse_transform method of your y-scaler. Note that you can do all this more concisely using a pipeline, as follows

from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.svm import SVR

pipeline = Pipeline([('scaler', StandardScaler()), ('estimator', SVR(kernel="rbf"))])

y_scaler = StandardScaler()
y_train = y_scaler.fit_transform(y_train)
pipeline.fit(x_train, y_train)
y_pred = y_scaler.inverse_transform(pipeline.predict(x_test))

Many would just scale the target globally and get away without too much overfitting. But you are doing good in not falling for this. AFAIK using a separate scaler for y data as shown in the code is the only way to go.

Answer 2

I know this question is old and the answer was correct at the time, but there is a scikit-learn method of doing this now.

http://scikit-learn.org/dev/modules/compose.html#transforming-target-in-regression

Answer 3

As others have mentioned already you ought to use inverse_transform() method to retrieve original data from its respective transformation applied prior. Another point to ponder is, why the need to transform the Target y_test, y_train if our intention is to predict for the Real Target 'y' values ? We might as well have in it in its original state during prediction.

Also (in Python 3.7.3, sklearn 0.20.3), when you standardize single columned rows like y_test, y_train like you have done above you inadvertently receive output as an numpy array which won't help in Dataframe operations;

eg:

When you do specify that your output should resemble a single columned Dataframe, you may encounter more issues;

eg:

SOLUTION : You will have to explicitly state that your Target Column Name/Index in a List using proper subset selection operators (.loc/.iloc).

eg:

NOTE: in live ML projects, Test Data is something that arrives in the future or collected live when your model is ready for tuning during productionizing stage.

A standardized training-test Features set like X_train, X_test help in easy comparison of Feature variation from mean as well as useful for Regularization and Principal Component Analysis techniques which mandate standardization of feature variables.