def setup(self, stage: Optional[str] = None):
dataloaders = sidechainnet.load(
casp_version=self.casp_version,
scn_dir=self.scn_dir,
with_pytorch="dataloaders",
)
print(dataloaders.keys()
) # ['train', 'train_eval', 'valid-10', ..., 'valid-90', 'test']
self.train = ScnDataset(dataloaders["train"].dataset,
self.train_max_len)
self.val = ScnDataset(dataloaders["valid-90"].dataset,
self.eval_max_len)
self.test = ScnDataset(dataloaders["test"].dataset, self.eval_max_len)
distances = torch.cdist(coords, coords, p=2)
boundaries = torch.linspace(2,
20,
steps=DISTOGRAM_BUCKETS,
device=coords.device)
discretized_distances = torch.bucketize(distances, boundaries[:-1])
discretized_distances.masked_fill_(~(mask[:, :, None] & mask[:, None, :]),
IGNORE_INDEX)
return discretized_distances
# get data
data = scn.load(casp_version=12,
thinning=30,
with_pytorch='dataloaders',
batch_size=1,
dynamic_batching=False,
return_masks=True)
data = iter(data['train'])
data_cond = lambda t: t[1].shape[1] < THRESHOLD_LENGTH
dl = cycle(data, data_cond)
# model
model = Alphafold2(dim=256, depth=1, heads=8, dim_head=64).to(DEVICE)
# optimizer
optim = Adam(model.parameters(), lr=LEARNING_RATE)
depth=2,
attend_self=True,
input_degrees=1,
output_degrees=2,
reduce_dim_out=True,
differentiable_coors=True,
num_neighbors=0,
attend_sparse_neighbors=True,
num_adj_degrees=2,
adj_dim=4,
num_degrees=2,
)
data = scn.load(casp_version=12,
thinning=30,
with_pytorch='dataloaders',
batch_size=BATCH_SIZE,
dynamic_batching=False)
# Add gaussian noise to the coords
# Testing the refinement algorithm
dl = cycle(data['train'])
optim = Adam(transformer.parameters(), lr=1e-4)
transformer = transformer.cuda()
for _ in range(10000):
for _ in range(GRADIENT_ACCUMULATE_EVERY):
batch = next(dl)
seqs, coords, masks = batch.seqs, batch.crds, batch.msks
seqs = seqs.cuda().argmax(dim=-1)
