A question about CV for 4th IFC

[gangliu-github]

I generated independent random sampling configurations from the harmonic PES at 300 K, following the 7.6 of the Tutorial. I want to obtain the 4th IFCs. But when I performed the CV, I found the RESIDUAL error was above 50%. I also found the strange information:
RESIDUAL (%): 68.1987
Number of non-zero HARMONIC FCs : 0
Number of non-zero ANHARM3 FCs : 0
Number of non-zero ANHARM4 FCs : 1
What is the problem? The files are uploaded.
cv.zip

[ttadano]

In cv.log, you can see that the L1 penalty alpha value is very large for the 2dn and 4th CV sets, while they are small for the 1st and 3rd CV sets. This implies that your training data contains some anomalies (e.g., unconverged results or unphysically high-energy structures).

[gangliu-github]

Thanks for your reply! Now I use the random structures from AIMD calculations, but the same problem also appears. Then, how to solve this problem? If I re-generate these random structures, may the problem be solved?