机器学习中常用的几种KERNEL

我看神经网络大多使用SIGMOID，其他非线性的除RBF和STRING之外的KERNEL也都应该很好用，这里给出一些相关介绍，有兴趣的话可以尝试

There are several commonly used kernel functions in machine learning and scientific computing:

- Linear kernel - Simple dot product between two vectors.

- Polynomial kernel - Models nonlinear relationships by degree.

- Radial basis function (RBF) kernel - Models similarity based on distance between points. Most popular kernel.

- Sigmoid kernel - Uses tanh function to transform data.

- Laplacian kernel - Based on the exponential of the negative absolute difference between points.

- Chi-square kernel - Uses chi-square statistic as a measure of similarity.

- ANOVA kernel - Uses ANOVA statistical technique to compute similarity.

- String kernels - Designed for text and sequences, like ngram matching.

Some common choices for kernel PCA include:

- RBF kernel - Works well on many datasets, only one parameter to tune.

- Laplacian kernel - Very similar to RBF, also a good general purpose kernel.- Polynomial kernel - Flexible but requires tuning degree hyperparameter.

- Sigmoid kernel - Can perform well but is not as commonly used.For a large dataset,

I would recommend starting with the RBF kernel first as it generally tends to perform well. The Laplacian kernel would be another safe option to try.

Polynomial and sigmoid may require more tuning. The linear kernel would not be ideal for nonlinear kernel PCA.

Make sure to subsample data if needed to experiment and fine-tune the kernel parameters. The optimal kernel ultimately depends on the dataset and problem at hand.