def __init__(self,
dim,
window,
activation='relu',
num_layers=2,
pooling_dim=1):
"""
:param dim: int
final dimension of the protein representation
:param activation:
non-linearity to apply to logits
:param window:
max size of grouped amino acids
:param num_layers: int
Number of convolution layers
:param pooling_dim: int
The dimension to be used in reducing protein segments to form a vector representation of the protein.
"""
super(ProteinCNN, self).__init__()
self.activation = get_activation_func(activation)
self.pooling_dim = pooling_dim
self.lin_kernels = nn.ModuleList([
nn.Linear(dim * window, dim * window)
for _ in range(num_layers - 1)
])
self.lin_kernels.append(nn.Linear(dim * window, dim))
def __init__(self, activation=None):
super(Layer, self).__init__()
if activation and isinstance(activation, str):
self.activation = get_activation_func(activation)
else:
self.activation = activation
def __init__(self, in_features, out_features, dprob, activation='relu'):
super(ResBlock, self).__init__()
self.activation = get_activation_func(activation)
self.net1 = nn.Sequential(nn.Linear(in_features, out_features),
nn.BatchNorm1d(out_features), nn.ReLU(),
nn.Dropout(dprob))
self.net2 = nn.Sequential(nn.Linear(out_features, out_features),
nn.BatchNorm1d(out_features), nn.ReLU(),
nn.Dropout(dprob))
self.bn_out = nn.BatchNorm1d(out_features)
def __init__(self,
num_fingerprints,
embedding_dim,
num_layers=3,
update='mean',
activation='relu'):
super(GraphNeuralNet, self).__init__()
self.output = update
self.activation = get_activation_func(activation)
self.embed_fingerprint = nn.Embedding(num_fingerprints,
int(embedding_dim))
self.W_gnn = nn.ModuleList([
nn.Linear(embedding_dim, embedding_dim) for _ in range(num_layers)
])
def __init__(self,
in_dim,
hidden_dim,
dropout,
num_layers=1,
bidrectional=False,
activation='relu',
batch_first=False):
super(ProteinRNN, self).__init__()
self.hidden_dim = int(hidden_dim)
self.num_layers = int(num_layers)
self.directions = max(1, int(bidrectional) + 1)
self.activation = get_activation_func(activation)
self.batch_first = batch_first
if num_layers == 1:
dropout = 0
self.model = nn.LSTM(input_size=int(in_dim),
hidden_size=int(self.hidden_dim),
num_layers=int(self.num_layers),
batch_first=batch_first,
dropout=dropout,
bidirectional=bidrectional)
def __init__(self,
protein_profile,
vocab_size,
embedding_dim,
activation='relu',
batch_first=False):
"""
:param protein_profile:
:param vocab_size: int
The number of ngrams formed (excluding padding index). An additional row is added to the end of the
embedding table constructed as a padding index.
:param embedding_dim:
:param activation:
:param batch_first:
"""
super(Prot2Vec, self).__init__()
self._batch_first = batch_first
self.protein_profile = protein_profile
self.embedding = nn.Embedding(vocab_size + 1,
int(embedding_dim),
padding_idx=vocab_size)
self.activation = get_activation_func(activation)
def __init__(self, activation='relu'):
super(SiameseNonlinearity, self).__init__()
self._func = get_activation_func(activation)
def __init__(self, dim1, dim2, out_dim, activation='relu'):
super(PreSiameseLinear, self).__init__()
self.linear1 = nn.Linear(dim1, out_features=out_dim)
self.linear2 = nn.Linear(dim2, out_features=out_dim)
self.activation = get_activation_func(activation)