Hi,
I am wondering whether there is a possibility to output the parameters of the list of iterations as a file. I’d like to follow how the percentage of, say class 2, evolves in time with increasing numbers of iterations.
It is also not clear to me whether the class numbers displayed in ‘Experiments’ and in ‘Launch’ are ordered in the same way. For example, in ‘Experiments’ I see that class 2 (not displayed as such but counted from the beginning of the displayed images) has 6.6% whereas at the very end of the list in ‘Launch’ it has has 9%…
Good question.
Right now there isn’t an easy way to get the class sizes as a function of iteration via command line (unless you know python and can use pymongo to read the cryoSPARC database directly).
The reason you see differences between the Experiments page and the Launch page is because: during ab-initio reconstruction we’re using the SGD algorithm which randomly samples images in each iteration. This means no iteration sees the entire dataset, but rather only a small subset. So the class sizes written in the log (Launch page) are percentages of the minibatch (random sample) in each iteration. On the other hand on the Experiments page, the class sizes are the actual percent of particles in the whole dataset that have been classified into each class.