halo light cones: reduce memory usage in prepare_sim#173
Conversation
abacusnbody/hod/prepare_sim.py
Outdated
| allpos[index_bounds], r=rad_outer, workers=nthread | ||
| ) | ||
| # this is the true number of randoms | ||
| for ind in np.arange(len(index_bounds)): |
There was a problem hiding this comment.
This should be for ind in range(len(index_bounds)), right? No need to construct the full array with np.arange.
There was a problem hiding this comment.
Yes, thanks for catching! Also changed it in one more place
| # repeat until condition satisfied | ||
| while count < len(index_bounds) * rand_final: | ||
| # generate randoms in L shape | ||
| randpos, randdist = gen_rand( |
There was a problem hiding this comment.
The test failure is due to this function calling np.random functions a different number of times than before, causing subsequent np.random calls to return different results. Maybe we should decouple this RNG with later RNG?
In other words, can this function be modified to accept an RNG object (created by np.random.default_rng(seed)) instead of using np.random directly? And then we reuse the RNG object for all calls to this function, and a different RNG object for other parts of the code?
There was a problem hiding this comment.
Makes sense! Included rng object set at newseed that is passed to that function -- hopefully it passes the tests now (though I worry that we also had a check for Menv so it might still fail because the values are different?)
There was a problem hiding this comment.
I don't think it will fix the tests, because previously the RNG for the randoms and Menv was coupled, and now we're decoupling it. But at least it will prevent the tests from breaking in the future.
This lets us give informative error messages that include the field name, and reduces repetition.
lgarrison
changed the title
But separately from the seed used for the rest of the randoms.
…utils into mem_prepare_sim Please enter a commit message to explain why this merge is necessary,
2 participants
Attempting a memory improvement of the compute environment script in
prepare_sim.py.Addressing issue #143