2023-04-06

x = e3nn.IrrepsArray("0e + 1o", jnp.array([1.0, 1.0, 1.0, 2.0]))
norm_activation(x, [None, jnp.tanh])

1x0e+1x1o [1.        0.8883856 0.8883856 1.7767712]

Changed

• e3nn.normalize_function now uses a deterministic (not pseudorandom) algorithm to compute the normalization factor.

Added

• normalize_act option to e3nn.scalar_activation and e3nn.gate. We can now turn the normalization off if we want to.
• e3nn.norm_activation as a new activation function.