Seed random seed

seed Application and usage in the model _zyrlia The blog of -CSDN Blog _ In depth learning seedSEEDseed In deep learning code, it is called random seed , Set up seed Because the initial weight parameters in the deep learning network model are usually initialized into random numbers . The local optimal solution obtained by using the gradient descent method is very sensitive to the selection of the initial location , Set up seed It is equivalent to specifying the initial random value . The method of setting random seeds can approximately and completely reproduce the author's open source deep learning code , The selection of random seeds can reduce the randomness of the algorithm results to a certain extent , That is, it is closer to the result of the original author, that is, generating random seeds means running experiments every time , The random numbers generated are the same. The following functions can be used in deep learning code def seed_all(seed): https://blog.csdn.net/qq_42766639/article/details/125061803seed In deep learning code, it is called random seed , Set up seed Because the initial weight parameters in the deep learning network model are usually initialized into random numbers . The local optimal solution obtained by using the gradient descent method is very sensitive to the selection of the initial location , Set up seed It is equivalent to specifying the initial random value .
The method of setting random seeds can approximately and completely reproduce the author's open source deep learning code , The selection of random seeds can reduce the randomness of the algorithm results to a certain extent , That is, it is closer to the result of the original author