Speed up PennyLane load time

[cvjjm]

Feature details

I have noticed that PennyLane import time is growing longer and longer and did a quick profiling session following this guide. The results look like this:

So, over 60% of the time is spent on importing qchem and most of that is attributable to openfermion. Now of course you can say that I should approach the openfermion people and tell them to speed up their load time, but clearly there are many programs that use pennylane, but don't need qchem and openfermion and in that case, even if those are installed, it would be nice if pennylane would start up fast.

Can qchem maybe be imported on demand?

Implementation

No response

How important would you say this feature is?

1: Not important. Would be nice to have.

Additional information

No response

[josh146]

Nice catch @cvjjm! Speeding up the import time would be a huge quality-of-life improvement.

Currently, we dynamically import qchem here, if the entry point exists:

pennylane/pennylane/__init__.py

Lines 140 to 142 in 854d4b4

	for entry in pkg_resources.iter_entry_points("pennylane.qchem"):
	if entry.name == "OpenFermion":
	qchem = entry.load()

but I'm sure we can somehow make this a 'lazy' import that is only loaded when first called.

[josh146]

@cvjjm I'm suddenly a bit confused. I can't recreate this behaviour locally; when I import pennylane as qml, I don't have OpenFermion automatically imported.

It seems we are already taking care of this in the qchem package:

pennylane/qchem/pennylane_qchem/qchem/openfermion.py

Lines 18 to 24 in 9687c13

	def __getattr__(name):
	"""Allow for lazy importing of openfermion"""
	try:
	return importlib.import_module("openfermion." + name)
	except ModuleNotFoundError:
	mod = importlib.import_module("openfermion")
	return getattr(mod, name)

Are you using an older version of qchem?

[cvjjm]

Indeed!! Looks like I picked a bad conda env for this test. This was with qchem 15.1 :-) Well, happy that at least this lead you to looking into making the whole qchem import lazy :-)