LinAlgError: SVD did not converge
Summary
Recently, Python was used to implement a mixed Gaussian distribution to generate training sets and test sets, and an error message SVD did not converge appeared in a part of the code for pseudo-inversion. After more than six hours of debugging, including searching for solutions to SVD did not converge on csdn, the problem is still not solved. Finally, I asked my classmates in the same lab for help and found that the generated samples would cause nan values in some intermediate operations. So, I tried to use a single Gaussian distribution to generate data. Sure enough, the error disappeared and the code ran smoothly. However, errors still occur when inverting data generated using a mixture of Gaussian distributions. Finally, I used matlab to generate mixed Gaussian distribution data (matlab has a ready-made mixed Gaussian distribution library that can be called).
Experience summary
If you encounter problems that are difficult to solve in python programming, you can consider them from the data perspective and observe whether there are nan values in the generated data.