Unexpected number of samples in Chainsail runs.

[Etjean]

Problem
Chainsail only dumps 1 sample out of 5, and this can't be changed by the user, which results in an unexpected number of samples in the end (if a user requests 10 000 samples, he will only get 2 000). This is set with the dump_step parameter, whose default value is 5.

Quote from Simeon, giving background for this (full slack thread here):

The subsampling is, I think, a leftover from my research, when I had issues with disk storage space (one simulation took around 500 Gb). I don't think there's a reason now to keep the subsampling.

Solution

• Change the default value of dump_step to 1.
• And optionally, add it as a configurable parameter in the frontend.

[simeoncarstens]

Note that when doing this, we might need to adapt the (hardcoded, I believe) settings for the DOS estimation, as it uses only a subset of samples and this number probably becomes 5x bigger if dump_step were set to 1.