As we know, in the research of machine learning or deep learning, training set and test set should be divided. The work of the training set is to train the model, while the function of the test set is to test the model on the trained model.However, I have seen some articles that test a model using the training set itself in order to demonstrate the effect of a model. Is there a problem with such logic?
No, you cannot use the training data for testing (model validity or performance). The basic principal is such that, you reserve a part of data for training and the remaining unseen/New-Data for testing purposes.
It is just a similar case of one's schooling levels. For example, you receiving education/training until certain level/Age, you than making decisions for the new events based on your previous/education/Training....and showing maturities. The new events/data may or may not be similar to that of training in new experience of life.
Muhammad Ali thank you for your answers ,but i have saw many papers do their research as i said，maybe they just want offer a method for us，but it can make a mistake for other.
Yes, the fair and conventional way is to split dat for training and testing purposes. In data analysis the training data is called known, whilst the remaining testing is called NEW/UNSEEN data that may be used for model/algorithm/method validation. For deep understanding, I suggest you the attached book.
- thank you very much! I will read this book for clear understanding. in fact i agree with what you said, but when i read some papers like what i said, i doubt if this logic is appropriate.But sometimes , to illustrate the appropriateness of the method, use training set to test the model can be used.
It may lead to overfitting where the model will fail to generalize to a new set outside the training set.
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training