How to ensure the consistency of the index， during training stage and prediction stage?

[paulxu1314]

lbe = LabelEncoder()
data[feat] = lbe.fit_transform(data[feat])
example:
in traing stage, feat_1:['a', 'b', 'c']
in prediction stage, feat_1:['a', 'c', 'd']

[zanshuxun]

As shown in the example, a label encoder is used to re-encode the features in the entire dataset data, and then data is divided into a train and test, so that the mapping relationship of the feature values ​​is consistent in the training set and the test set.

DeepCTR-Torch/examples/run_classification_criteo.py

Line 24 in f685425

data[feat] = lbe.fit_transform(data[feat])