[Test Failure] test_ode.py::test_random_network

[Madelyn Shapiro]

Description:

./tests/test_ode.py::test_random_network Failed: [undefined]TypeError: GeneralNetworkedODE.ode_equations() missing 1 required positional argument: 'u'

Full Testing Log

TestFailureMessage (click to expand)

./tests/test_ode.py::test_random_network Failed: [undefined]TypeError: GeneralNetworkedODE.ode_equations() missing 1 required positional argument: 'u'
Falsifying example: test_random_network(
    nAgents=1,
    nCouplings=0,
    nu=0,
    batchsize=1,
    bias='additive',
    system=neuromancer.dynamics.ode.GeneralNetworkedODE,
)
@given(st.integers(1, 100),
>          st.integers(0, 1000),
           st.integers(0, 100),
           st.integers(1, 500),
           st.sampled_from(bias),
           st.sampled_from(ode_networked_systems))

tests/test_ode.py:52: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
tests/test_ode.py:80: in test_random_network
    y = ode(x)
.conda/lib/python3.10/site-packages/torch/nn/modules/module.py:1501: in _call_impl
    return forward_call(*args, **kwargs)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = GeneralNetworkedODE(
  (agents): ModuleList(
    (0): SourceSink()
  )
  (couplings): ModuleList()
)
x = tensor([[0.3018]]), args = ()

    def forward(self, x, *args):
        assert len(x.shape) == 2
>       return self.ode_equations(x, *args)
E       TypeError: GeneralNetworkedODE.ode_equations() missing 1 required positional argument: 'u'
E       Falsifying example: test_random_network(
E           nAgents=1,
E           nCouplings=0,
E           nu=0,
E           batchsize=1,
E           bias='additive',
E           system=neuromancer.dynamics.ode.GeneralNetworkedODE,
E       )

src/neuromancer/dynamics/ode.py:24: TypeError

[Madelyn Shapiro]

Context

The 1.4.1 release (#63):

• modified ode.py API for non-autonomous systems (DuffingParam, TwoTankParam, LorenzControl, CSTR_Param, & VanDerPolControl) so that the ode_equations method accepts x and u as two separate mandatory arguments instead of a single stacked tensor.
• test coverage in test_ode.py::test_ode_nonauto_param_shape was updated to use the new API.

In response to issue #76, hotfix 22ebe07:

• applied the same API change to networked systems (GeneralNetworkedODE) to support RC networks.
• test coverage in test_ode.py::test_random_network was not updated.

Main problem: Backwards compatibility of ODEs API in 1.4.1

For the blocks interface, the Block base class handles either input format (stacked x or separate x, *args.) The dynamics module needs equivalent handling to keep non-autonomous and networked systems backwards-compatible.

Proposed Fix

• Update function signature for ode_equations so that u is an optional arg, default None
• For non-autonomous systems: If u is None and x.shape[-1] == self.nx + self.nu: split into separate x, u tensors and issue a warning to the user about the API update before proceeding with the rest of the function.
• For networked systems: wrap a conditional around the concatenation of x, u to check if u is not None

Status

PR in progress

TODO:

• Implement proposed fix described above
• Update test coverage for both current API + backwards compatibility

[Ján Drgoňa]

I fixed this issue by supporting *args inputs in ode_equations. The proposed solution is in the development branch.

[Ján Drgoňa]

resolver in the most recent PR