Validate point-wise torch activation in eMLP and iMLP

[Copilot]

eMLP and iMLP silently accepted any torch.nn.Module as activation, including non-point-wise modules that break equivariance. Add an elegant, forward-compatible guard that rejects anything that isn't a torch point-wise activation.

Changes

• _is_torch_pointwise_activation(module) — new module-level helper in emlp.py. Checks type(module).__module__ == "torch.nn.modules.activation", which covers every current and future activation in PyTorch's dedicated activation namespace without any manual enumeration.
• eMLP.__init__ — replaces bare assert isinstance(activation, torch.nn.Module) with a ValueError using the helper.
• iMLP.__init__ — adds the same guard (previously no activation type check existed).

# Accepted — any torch.nn activation, including future ones
eMLP(..., activation=torch.nn.SiLU())

# Raises ValueError: "Expected a point-wise non-linearity from `torch.nn` ..."
eMLP(..., activation=torch.nn.Sequential(torch.nn.ReLU()))
eMLP(..., activation=torch.nn.Linear(4, 4))

• Tests — two new parametrized cases: one asserting several torch activations (ReLU, GELU, SiLU, Tanh, ELU) are accepted; one asserting non-activation modules raise ValueError with a message containing the offending type name.

Original prompt

in the eMLP and iMLP models, I would like an easy way to understand if the user passed any of the point-wise non-linearities used by torch. Ideally an elegant check, such that I dont need to handle any new activation funciton added.

Can you create a PR adding this feature ?

Without bothering me

The user has attached the following file paths as relevant context:

• pyproject.toml

Created from VS Code.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.