manodeep / astro3d

Python package for ASTRO 3D

astro3D

Python package for the Genesis simulations from ASTRO 3D

Documenation is hosted on ReadTheDocs

astro3d's Issues

hostHaloID is -1 for FoF Halo

(Just a note for myself so I can remember)

Looking at the new Genesis tree format, hostHaloID (the analog for FirstHaloInFOFgroup) has been changed to be -1 for a FoF Halo. A core assumption in treefrog_to_lhalo.py is that this instead points to itself. This is causing issues with (at least currently) infinite loops after fixing the flybys.

Will need to check how this assumption propagates through the code and fix it up.

MemoryError for get_halos_per_forest

When trying to convert the temporal indices to the tree local ones (function convert_indices in convert_indices.py) for the 2048^3 simulation I hit a MemoryError when calling get_halos_per_forest from the common.py module.

Traceback (most recent call last):
  File "run_convert_indices.py", line 30, in <module>
    ID_fields, index_mult_factor)
  File "/home/jseiler/.local/lib/python3.6/site-packages/astro3D-0.0.0.dev2-py3.6.egg/astro3D/genesis/utils/convert_indices.py", line 82, in convert_indices
    forestID_field)
  File "/home/jseiler/.local/lib/python3.6/site-packages/astro3D-0.0.0.dev2-py3.6.egg/astro3D/genesis/utils/common.py", line 298, in get_halos_per_forest
    forests_binned = np.bincount(halo_forestids)
MemoryError

I've added some print statements and I'm baffled by this error. The arrays that it's using aren't excessive in size. The only thing that could be happening is that the open HDF5 file is chewing up all the RAM? However, I've run this through the queue on OzStar with 200GB of RAM requested and still hit the same issue...

@manodeep : If you have time during the retreat, could you look into this? The dev branch has all the updated filepaths. You should be able to run these with the scripts in the run_scripts directory; in particular, the run_convert_indices.py file.

HDF5 Metadata Issue

I ran into this issue previously. In essence, we save the Number of Halos per tree in a header attribute. However, for large simulations the number of trees is extraordinarily large, the datasize of the attribute exceeds the 64kb limit.

Possible solutions are:

1. Write NHalosPerTree as a data object instead of an attribute (i.e., make it a group that can be accessed via a key). Will need to investigate the reprecussions of this, the least of which would be to adjust the SAGE HDF5 reader.

2. Write to more files so a single file doesn't exceed 64kb. This isn't scalable so probably not the best solution...

For now, I'm just going to write only to a binary file.