Altair RISE

A program to recognize and reward our most engaged community members

Nominate Yourself Now!

hm.Collection is very slow

Pierre_20718

Hi everyone,
I need to batch-process 450 different regions. For each region, I generate from node ids an RBE2.
With a Python solution (hm.Collection), this operation is very slow : for a sample of "only" 85 regions, the process takes 160 seconds (around 3 minutes) and usesa lot of RAM.
However, using TCL marks in Python (with hw.evalTcl) in an equivalent source code takes only 3 seconds for this 85 regions !

Why this difference ?
To accelerate my "full-Python" code, I think I need to stop creating a collection on each iteration of the loop. Instead, I should empty the list of node ids and fill it once again with new ids for the next item in the loop. How can I do this, efficiently ?

Thanks

Find more posts tagged with

HyperMesh

Scripts

Python

v2025.1

Accepted answers

Michal Stefuca

Hi @Pierre_20718

Thank you for sharing the code.

Assuming nodeids is a Python list (or numpy array) of integers, can you please try swapping

col_tmp = hm.Collection(model, ent.Node, f"id={list2listr(nodeids)}")

for

col_tmp = hm.Collection(model, ent.Node, nodeids)

and see if the performance improves?

In general, we still have few areas in the pipeline where we are trying to improve the performance of Python API comparing to Tcl. Certain operations are naturally "heavier" in Python as we are dealing with generating/handling objects in the background, but as workflows change and infrastructure develops, this should not pose performance drawbacks.

Out of curiosity, which version of HyperMesh are you using?

Thanks,
Michal

All comments

Michal Stefuca

Hi @Pierre_20718

Could you please share a snippet of your code so we can better understand the exact usage of hm.Collection? What entities are you selecting, what criteria is used, how is the collection object subsequently used in the code, etc.

Best regards,

Michal Stefuca
Automation & Customization

Pierre_20718

Thanks for your message, @Michal Stefuca ,

I wrote this code:

def list2listr(listin, sep:str=","):

    items = listin if isinstance(listin, (list, tuple)) else [listin]

    return sep.join(map(str, items))




# PythonHM way


for i, (geomids, nodeids) in enumerate(region_zip):
        # nodeids is list of integers
        # geomids is not used here


	col_tmp = hm.Collection(model, ent.Node, f"id={list2listr(nodeids)}")

	model.rigidlinkinodecalandcreate(node_collection1=col_tmp, dof=123456, set_id=0, attach_set_option=0)

#



# TCL way

for i, (geomids, nodeids) in enumerate(region_zip):

	TCL_reset_nodes = "*clearmark node 1"

	TCL_set_nodes = '*createmark nodes 1 "by id only" '+list2listr(nodeids, sep=" ")

	TCL_creatRigid = "*rigidlinkinodecalandcreate 1 0 0 123456"


	TCL = "\n".join([TCL_reset_nodes, TCL_set_nodes, TCL_creatRigid])

	hw.evalTcl(TCL)

#

I also tried to benchmark both API on another function:

_nb = 100000

L = [None,]*_nb 


print("PyHM")

start_time = time.time()

for i in range(_nb):

   L[i] = model.hm_latestentityid(ent.Element, 0)[1].entity

#

cur_time = time.time()

print("--- %s seconds ---" % (cur_time - start_time))





print("TCL")


L = [None,]*_nb 

start_time = time.time()

for i in range(_nb):

   L[i] = hw.evalTcl("hm_latestentityid elems")

#

cur_time = time.time()

print("--- %s seconds ---" % (cur_time - start_time))

And I get for this test:

PyHM
--- 1.9589903354644775 seconds ---
TCL
--- 0.09697127342224121 seconds ---

Michal Stefuca

Hi @Pierre_20718

Thank you for sharing the code.

Assuming nodeids is a Python list (or numpy array) of integers, can you please try swapping

col_tmp = hm.Collection(model, ent.Node, f"id={list2listr(nodeids)}")

for

col_tmp = hm.Collection(model, ent.Node, nodeids)

and see if the performance improves?

Out of curiosity, which version of HyperMesh are you using?

Thanks,
Michal

Pierre_20718

Thanks @Michal Stefuca !

Using a list of integers (nodeids) and not a string give similar times (2.49 seconds) to TCL (2.58s) !

col_tmp = hm.Collection(model, ent.Node, nodeids)

I didn't tried it earlier because in https://help.altair.com/hwdesktop/pythonapi/hypermesh/hm.html#id4 , it seems only for Parts ;)

I use HM 2025.1