def test_channels(self):
tau = SO3Tau([3, 2, 1])
assert tau.channels == None
tau = SO3Tau([3] * 2)
assert tau.channels == 3
def __init__(self,
max_sh,
basis_set,
num_channels,
mix=False,
device=torch.device('cpu'),
dtype=torch.float):
super(RadPolyTrig, self).__init__()
trig_basis, rpow = basis_set
self.rpow = rpow
self.max_sh = max_sh
assert (trig_basis >= 0 and rpow >= 0)
self.num_rad = (trig_basis + 1) * (rpow + 1)
self.num_channels = num_channels
# This instantiates a set of functions sin(2*pi*n*x/a), cos(2*pi*n*x/a) with a=1.
self.scales = torch.cat(
[torch.arange(trig_basis + 1),
torch.arange(trig_basis + 1)]).view(1, 1, 1, -1).to(device=device,
dtype=dtype)
self.phases = torch.cat([
torch.zeros(trig_basis + 1), pi / 2 * torch.ones(trig_basis + 1)
]).view(1, 1, 1, -1).to(device=device, dtype=dtype)
# This avoids the sin(0*r + 0) = 0 part from wasting computations.
self.phases[0, 0, 0, 0] = pi / 2
# Now, make the above learnable
self.scales = nn.Parameter(self.scales)
self.phases = nn.Parameter(self.phases)
# If desired, mix the radial components to a desired shape
self.mix = mix
if (mix == 'cplx') or (mix is True):
self.linear = nn.ModuleList([
nn.Linear(2 * self.num_rad,
2 * self.num_channels).to(device=device, dtype=dtype)
for _ in range(max_sh + 1)
])
self.tau = SO3Tau((num_channels, ) * (max_sh + 1))
elif mix == 'real':
self.linear = nn.ModuleList([
nn.Linear(2 * self.num_rad,
self.num_channels).to(device=device, dtype=dtype)
for _ in range(max_sh + 1)
])
self.tau = SO3Tau((num_channels, ) * (max_sh + 1))
elif (mix == 'none') or (mix is False):
self.linear = None
self.tau = SO3Tau((self.num_rad, ) * (max_sh + 1))
else:
raise ValueError(
'Can only specify mix = real, cplx, or none! {}'.format(mix))
self.zero = torch.tensor(0, device=device, dtype=dtype)
def test_init(self, maxl, num_channels):
tau = SO3Tau([num_channels] * (maxl + 1))
assert type(tau) == SO3Tau
assert list(tau) == [num_channels] * (maxl + 1)
tau = SO3Tau(tau)
assert type(tau) == SO3Tau
assert list(tau) == [num_channels] * (maxl + 1)
def test_SO3Weight_init(self, maxl, channels1, channels2):
tau1_list = SO3Tau([channels1] * (maxl + 1))
tau2_list = SO3Tau([channels2] * (maxl + 1))
weight_list = rand_weight_list(tau1_list, tau2_list)
weight = SO3Weight(weight_list)
assert isinstance(weight, SO3Weight)
assert weight.tau_in == SO3Tau(tau1_list)
assert weight.tau_out == SO3Tau(tau2_list)
def __init__(self, taus_in, maxl=None):
super().__init__()
self.taus_in = taus_in = [SO3Tau(tau) for tau in taus_in if tau]
if maxl is None:
maxl = max([tau.maxl for tau in taus_in])
self.maxl = maxl
self.tau_out = reduce(lambda x, y: x & y, taus_in)[:self.maxl + 1]
def __init__(self, tau_in=None, cat=True, device=None, dtype=None):
super().__init__(device=device, dtype=dtype)
self.tau_in = tau_in
self.cat = cat
if self.tau_in is not None:
if cat:
self.tau = SO3Tau([sum(tau_in)] * len(tau_in))
else:
self.tau = SO3Tau([t for t in tau_in])
self.signs = [
torch.tensor(-1.).pow(torch.arange(-ell, ell + 1).float()).to(
device=self.device, dtype=self.dtype).unsqueeze(-1)
for ell in range(len(tau_in) + 1)
]
self.conj = torch.tensor([1., -1.]).to(device=self.device,
dtype=self.dtype)
else:
self.tau = None
self.signs = None
def test_cat(self):
tau1 = SO3Tau([1, 2, 3])
tau2 = SO3Tau([1, 1])
tau3 = SO3Tau([0, 0, 2])
tau = SO3Tau.cat([tau1, tau2])
assert list(tau) == [2, 3, 3]
assert type(tau) == SO3Tau
print(tau)
tau = (tau1 & tau2)
assert list(tau) == [2, 3, 3]
tau1 &= tau2
assert list(tau1) == [2, 3, 3]
tau123 = (tau1 & tau2) & tau3
assert SO3Tau.cat([tau1, tau2, tau3]) == tau123
def __init__(self,
tau_in,
tau_out,
real=False,
weight_init='randn',
gain=1,
device=None,
dtype=None):
super().__init__(device=device, dtype=dtype)
tau_in = SO3Tau(tau_in)
tau_out = SO3Tau(tau_out) if type(tau_out) is not int else tau_out
# Allow one to set the output tau to a pre-specified number of output channels.
if type(tau_out) is int:
tau_out = [tau_out] * len(tau_in)
self.tau_in = SO3Tau(tau_in)
self.tau_out = SO3Tau(tau_out)
self.real = real
if weight_init is 'randn':
weights = SO3Weight.randn(self.tau_in,
self.tau_out,
device=device,
dtype=dtype)
elif weight_init is 'rand':
weights = SO3Weight.rand(self.tau_in,
self.tau_out,
device=device,
dtype=dtype)
weights = 2 * weights - 1
else:
raise NotImplementedError(
'weight_init can only be randn or rand for now')
gain = [gain / max(shape) for shape in weights.shapes]
weights = gain * weights
self.weights = weights.as_parameter()
def cg_product_tau(tau1, tau2, maxl=inf):
"""
Calulate output multiplicity of the CG Product of two SO3 Vectors
given the multiplicty of two input SO3 Vectors.
Parameters
----------
tau1 : :class:`list` of :class:`int`, :class:`SO3Tau`.
Multiplicity of first representation.
tau2 : :class:`list` of :class:`int`, :class:`SO3Tau`.
Multiplicity of second representation.
maxl : :class:`int`
Largest weight to include in CG Product.
Return
------
tau : :class:`SO3Tau`
Multiplicity of output representation.
"""
tau1 = SO3Tau(tau1)
tau2 = SO3Tau(tau2)
L1, L2 = tau1.maxl, tau2.maxl
L = min(L1 + L2, maxl)
tau = [0] * (L + 1)
for l1 in range(L1 + 1):
for l2 in range(L2 + 1):
lmin, lmax = abs(l1 - l2), min(l1 + l2, maxl)
for l in range(lmin, lmax + 1):
tau[l] += tau1[l1]
return SO3Tau(tau)
def test_add(self):
tau1 = SO3Tau([1, 2, 3])
tau2 = SO3Tau([1, 1])
tau = tau1 + tau2
assert type(tau) == SO3Tau
assert list(tau) == list(tau1) + list(tau2)
tau = tuple(tau1) + tau2
assert type(tau) == SO3Tau
assert list(tau) == list(tau1) + list(tau2)
tau = list(tau1) + tau2
assert type(tau) == SO3Tau
assert list(tau) == list(tau1) + list(tau2)
tau1p = SO3Tau(tau1)
tau1p += tau2
assert type(tau1p) == SO3Tau
assert list(tau1p) == list(tau1) + list(tau2)
tau = sum([SO3Tau([3, 2, 1]), [1], (2, 3)])
assert type(tau) == SO3Tau
assert list(tau) == [3, 2, 1, 1, 2, 3]
def test_CGProduct(self, batch, maxl1, maxl2, maxl, channels):
maxl_all = max(maxl1, maxl2, maxl)
D, R, _ = rot.gen_rot(maxl_all)
cg_dict = CGDict(maxl=maxl_all, dtype=torch.double)
cg_prod = CGProduct(maxl=maxl, dtype=torch.double, cg_dict=cg_dict)
tau1 = SO3Tau([channels] * (maxl1 + 1))
tau2 = SO3Tau([channels] * (maxl2 + 1))
vec1 = SO3Vec.randn(tau1, batch, dtype=torch.double)
vec2 = SO3Vec.randn(tau2, batch, dtype=torch.double)
vec1i = vec1.apply_wigner(D, dir='left')
vec2i = vec2.apply_wigner(D, dir='left')
vec_prod = cg_prod(vec1, vec2)
veci_prod = cg_prod(vec1i, vec2i)
vecf_prod = vec_prod.apply_wigner(D, dir='left')
# diff = (sph_harmsr - sph_harmsd).abs()
diff = [(p1 - p2).abs().max() for p1, p2 in zip(veci_prod, vecf_prod)]
assert all([d < 1e-6 for d in diff])
def __init__(self,
taus_in,
tau_out,
maxl=None,
real=False,
weight_init='randn',
gain=1,
device=None,
dtype=None):
super().__init__(device=device, dtype=dtype)
self.cat_reps = CatReps(taus_in, maxl=maxl)
self.mix_reps = MixReps(self.cat_reps.tau,
tau_out,
real=real,
weight_init=weight_init,
gain=gain,
device=device,
dtype=dtype)
self.taus_in = taus_in
self.tau_out = SO3Tau(self.mix_reps)
def __init__(
self,
observation_space: ObservationSpace,
action_space: ActionSpace,
min_max_distance: Tuple[float, float],
network_width: int,
maxl: int,
num_cg_levels: int,
num_channels_hidden: int,
num_channels_per_element: int,
num_gaussians: int,
bag_scale: int,
beta: Optional[float] = None,
device=None,
):
super().__init__(observation_space, action_space)
self.device = device
self.dtype = torch.float
self.zs = self.observation_space.zs
self.zs_tensor = torch.tensor(self.zs, dtype=self.dtype, device=self.device)
self.min_distance, self.max_distance = min_max_distance
assert self.min_distance < self.max_distance
self.beta = beta
self.max_sh = maxl
self.num_cg_levels = num_cg_levels
self.num_channels_hidden = num_channels_hidden
self.num_channels_per_element = num_channels_per_element
self.num_gaussians = num_gaussians
self.num_channels_out = len(self.zs) * self.num_channels_per_element
self.channel_offsets = torch.arange(start=0,
end=self.num_channels_per_element,
dtype=torch.long,
device=self.device).unsqueeze(0)
self.cg_dict = CGDict(maxl=self.max_sh, device=self.device, dtype=self.dtype)
self.cg_model = Cormorant(
maxl=self.max_sh, # Cutoff in CG operations (default: [3])
max_sh=self.max_sh, # Number of spherical harmonic powers to use (default: [3])
num_cg_levels=self.num_cg_levels, # Number of CG levels (default: 4)
num_channels=[self.num_channels_hidden] * self.num_cg_levels + [self.num_channels_out],
num_species=len(self.zs),
cutoff_type=['soft'], # Types of cutoffs to include
hard_cut_rad=min(self.max_distance, 2.1), # Radius of hard cutoff (in AA)
soft_cut_rad=min(self.max_distance, 2.1), # Radius of soft cutoff (in AA)
soft_cut_width=0.2, # Width of SOFT cutoff in Angstroms (default: 0.2)
weight_init='rand', # Weight initialization function to use (default: rand)
level_gain=[10.0], # Gain at each level (default: [10.])
charge_power=2, # Maximum power to take in one-hot (default: 2)
basis_set=[3, 3], # Use gaussian mask instead of sigmoid mask.
charge_scale=max(self.zs),
bag_scale=bag_scale,
device=self.device,
dtype=self.dtype,
cg_dict=self.cg_dict,
)
self.cg_mix = CormorantMixer(
tau_in=SO3Tau([self.num_channels_per_element] * (self.max_sh + 1)),
tau_other=SO3Tau([self.num_channels_per_element]),
maxl=self.max_sh,
num_channels=self.num_channels_per_element,
level_gain=10.0,
weight_init='rand',
device=self.device,
dtype=self.dtype,
cg_dict=self.cg_dict,
)
self.sph_harms = SphericalHarmonics(maxl=self.max_sh,
conj=False,
sh_norm='qm',
device=self.device,
dtype=self.dtype,
cg_dict=self.cg_dict)
self.atomic_scalars = AtomicScalars(maxl=self.max_sh, full_scalars=True, device=self.device, dtype=self.dtype)
self.num_latent = self.atomic_scalars.get_output_dim(self.num_channels_out)
self.num_latent_element = self.atomic_scalars.get_output_dim(self.num_channels_per_element)
# Focus
self.phi_focus = MLP(
input_dim=self.num_latent,
output_dims=(network_width, 1),
)
# Element
self.phi_element = MLP(
input_dim=self.num_latent,
output_dims=(network_width, len(self.zs)),
)
# Distance: Gaussian Mixture Model
self.phi_d = MLP(
input_dim=self.num_latent_element,
output_dims=(network_width, 2 * self.num_gaussians),
)
self.pad_zeros = torch.nn.ConstantPad1d(padding=(0, 1), value=0.0) # Pad with one 0.0 to the right
self.distance_half_width = torch.tensor((self.max_distance - self.min_distance) / 2,
dtype=self.dtype,
device=self.device)
self.distance_center = torch.tensor((self.min_distance + self.max_distance) / 2,
dtype=self.dtype,
device=self.device)
self.distance_log_stds = torch.nn.Parameter(torch.log(
torch.tensor([0.1] * self.num_gaussians, dtype=self.dtype, device=self.device)),
requires_grad=True) # (gaussians, )
# Value function
self.phi_trans = MLP(
input_dim=self.num_latent,
output_dims=(network_width, network_width),
)
self.phi_v = MLP(
input_dim=network_width,
output_dims=(network_width, 1),
)
self.to(self.device)
def tau(self):
return SO3Tau([])
def test_apply_euler(self, batch, channels, maxl):
tau = SO3Tau([channels] * (maxl + 1))
vec = SO3Vec.rand(batch, tau, dtype=torch.double)
wigner = SO3WignerD.euler(maxl, dtype=torch.double)
so3_torch.apply_wigner(wigner, vec)
def tau(self):
return SO3Tau([self.channels_out])