Tests the geometric_product() primitive for correctness (that is, consistency with the clifford library).

Tests the geometric_product() primitive for equivariance.

Tests the multivector reverse for equivariance.

Tests the grade_project() primitive for equivariance.

Tests an identity map for equivariance (testing the test).

Tests the innner_product() primitive for equivariance.

Tests the join() primitive for equivariance.

Tests the join_norm() primitive for correctness.

Tests the equi_linear() primitive for equivariance.

Tests the equi_linear() primitive for equivariance.

Tests the initialization of `EquiLinear`. The goal is that independent of the channel size, inputs with variance 1 are mapped to outputs with

Tests that the EquiLinear layer indeed describes a linear map (when the bias is deactivated). Checks that `f(x + rescaling * y) = f(x) + rescalin

Tests that minimum_autocast_precision() casts inputs correctly

Tests that minimum_autocast_precision() casts outputs correctly

Tests whether the conversion between lists of clifford.Multivectors and torch.Tensors is cycle-consistent.

Checks that norm() is consistent with inner_product().

Tests the norm() primitive for equivariance.

Computes the norm of a pure vector and compares against a known result.

Verifies contract_path can be called with shapes. Note that we test upstream functionality here, which should not be necessary. However, with

Tests the outer_product() primitive for correctness (that is, consistency with the clifford library).

Tests the outer_product() primitive for equivariance.

Tests pga_attention() for Pin equivariance.

Tests that with the right weights, pga_attention attends to nearby points.

Tests that outputs of pga_attention() have correct shape.

Tests the pin_invariants() primitive for equivariance.

Tests whether plane embeddings into multivectors are cycle consistent.

Tests whether 3D point embeddings into multivectors are cycle consistent.

Tests whether 3D points extracted from multivector are invariant under rescalings of the multivector.

Tests whether reflection embeddings into multivectors are cycle consistent.

Tests whether reflection embeddings act on oriented plane embeddings as expected.

Tests whether the embedding of a reflection along the plane normal to a given coordinate axis acts on point embeddings as expected.

Tests whether reflections act on pseudoscalars as expected.

Tests whether reflections act on scalars as expected.

Test whether model can successfully regress on a dataset data to almost zero train error.

Tests the multivector reverse for equivariance.

Tests whether rotation embeddings into multivectors are cycle consistent.

Tests whether applying rotations to planes via matrix-vector multiplication equals rotation via geometric product.

Tests whether applying rotations to points via matrix-vector multiplication equals rotation via geometric product.

Tests whether rotations act on pseudoscalars as expected.

Tests whether rotations act on scalars as expected.

Tests scalar_attention() for Pin equivariance.

Tests that outputs of scalar_attention() have correct shape.

Tests ScalarGatedNonlinearity() for equivariance.

Test to_nd.

Verifies that the global switch for gatr_einsum works as expected.

Tests the output shape of BaselineTransformer.

Tests whether translation embeddings into multivectors are cycle consistent.

Tests whether translation embeddings act on oriented plane embeddings as expected.

Tests whether translation embeddings act on point embeddings as expected.

Tests whether translations act on pseudoscalars as expected.

Tests whether translations act on scalars as expected.