Temporarily pin autograd < 1.6, fix builds

[agriyakhetarpal]

Description

Adds an upper-bound constraint to autograd's version, in order to fix builds for Windows and ARM M-series macOS, for more details see HIPS/autograd#601

Update: fixed after a 1.6.1 patch release
Fixes # (issue)

Type of change

Please add a line in the relevant section of CHANGELOG.md to document the change (include PR #) - note reverse order of PR #s. If necessary, also add to the list of breaking changes.

• New feature (non-breaking change which adds functionality)
• Optimization (back-end change that speeds up the code)
• Bug fix (non-breaking change which fixes an issue)

Key checklist:

• No style issues: $ pre-commit run (see CONTRIBUTING.md for how to set this up to run automatically when committing locally, in just two lines of code)
• All tests pass: $ python run-tests.py --all
• The documentation builds: $ python run-tests.py --doctest

You can run unit and doctests together at once, using $ python run-tests.py --quick.

Further checks:

• Code is commented, particularly in hard-to-understand areas
• Tests added that prove fix is effective or that feature works

[valentinsulzer]

We can probably just remove autograd as a dependency, we don't use it anywhere (only in one test specifically just for autograd)

[agriyakhetarpal]

Good idea, since we can fall back to using derivative instead of autograd when differentiating. Do you want me to try doing this in a further PR?

[valentinsulzer]

Yes please. I don't think the fall back to derivative will work, that is specifically for interpolants. But we never need the fallback in practice

[agriyakhetarpal]

The only place where we use it is here:

PyBaMM/pybamm/expression_tree/functions.py

Lines 94 to 122 in bf85811

	def _function_diff(self, children, idx):
	"""
	Derivative with respect to child number 'idx'.
	See :meth:`pybamm.Symbol._diff()`.
	"""
	# Store differentiated function, needed in case we want to convert to CasADi
	if self.derivative == "autograd":
	return Function(
	autograd.elementwise_grad(self.function, idx),
	*children,
	differentiated_function=self.function,
	)
	elif self.derivative == "derivative":
	if len(children) > 1:
	raise ValueError(
	"""
	differentiation using '.derivative()' not implemented for functions
	with more than one child
	"""
	)
	else:
	# keep using "derivative" as derivative
	return pybamm.Function(
	self.function.derivative(),
	*children,
	derivative="derivative",
	differentiated_function=self.function,
	)

Edit: great, opening a PR soon, but I would need some help though

[agriyakhetarpal]

@tinosulzer I thought of some more discussion before I start doing it concretely. Using "derivative" instead of "autograd" did not work, as you said above. Five tests from test_expression_tree.test_functions fail locally when I remove autograd as an option, and it is the error that comes up when len(children) > 1:

ValueError: differentiation using '.derivative()' not implemented for functions with more than one child

should I remove those tests? Debugging this could be slightly outside my forte

[valentinsulzer]

I see, let's keep this as is then, and just make autograd an optional dependency as is already being done in #3044

[agriyakhetarpal]

Sure, closing this since the build issue was resolved with autograd 1.6.2