File size: 5,860 Bytes
65bd8af |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
from torch import nn
from .modules import *
import pdb
class ConvLayer(nn.Module):
def __init__(self, in_channels, out_channels, kernel_size, padding, dilation):
super(ConvLayer, self).__init__()
self.conv = nn.Conv1d(in_channels, out_channels, kernel_size, padding=padding, dilation=dilation)
self.relu = nn.ReLU()
def forward(self, x):
out = self.conv(x)
out = self.relu(out)
return out
class DilatedCNN(nn.Module):
def __init__(self, d_model, d_hidden):
super(DilatedCNN, self).__init__()
self.first_ = nn.ModuleList()
self.second_ = nn.ModuleList()
self.third_ = nn.ModuleList()
dilation_tuple = (1, 2, 3)
dim_in_tuple = (d_model, d_hidden, d_hidden)
dim_out_tuple = (d_hidden, d_hidden, d_hidden)
for i, dilation_rate in enumerate(dilation_tuple):
self.first_.append(ConvLayer(dim_in_tuple[i], dim_out_tuple[i], kernel_size=3, padding=dilation_rate,
dilation=dilation_rate))
for i, dilation_rate in enumerate(dilation_tuple):
self.second_.append(ConvLayer(dim_in_tuple[i], dim_out_tuple[i], kernel_size=5, padding=2*dilation_rate,
dilation=dilation_rate))
for i, dilation_rate in enumerate(dilation_tuple):
self.third_.append(ConvLayer(dim_in_tuple[i], dim_out_tuple[i], kernel_size=7, padding=3*dilation_rate,
dilation=dilation_rate))
def forward(self, protein_seq_enc):
# pdb.set_trace()
protein_seq_enc = protein_seq_enc.transpose(1, 2) # protein_seq_enc's shape: B*L*d_model -> B*d_model*L
first_embedding = protein_seq_enc
second_embedding = protein_seq_enc
third_embedding = protein_seq_enc
for i in range(len(self.first_)):
first_embedding = self.first_[i](first_embedding)
for i in range(len(self.second_)):
second_embedding = self.second_[i](second_embedding)
for i in range(len(self.third_)):
third_embedding = self.third_[i](third_embedding)
# pdb.set_trace()
protein_seq_enc = first_embedding + second_embedding + third_embedding
return protein_seq_enc.transpose(1, 2)
class ReciprocalLayerwithCNN(nn.Module):
def __init__(self, d_model, d_inner, d_hidden, n_head, d_k, d_v):
super().__init__()
self.cnn = DilatedCNN(d_model, d_hidden)
self.sequence_attention_layer = MultiHeadAttentionSequence(n_head, d_hidden,
d_k, d_v)
self.protein_attention_layer = MultiHeadAttentionSequence(n_head, d_hidden,
d_k, d_v)
self.reciprocal_attention_layer = MultiHeadAttentionReciprocal(n_head, d_hidden,
d_k, d_v)
self.ffn_seq = FFN(d_hidden, d_inner)
self.ffn_protein = FFN(d_hidden, d_inner)
def forward(self, sequence_enc, protein_seq_enc):
# pdb.set_trace() # protein_seq_enc.shape = B * L * d_model
protein_seq_enc = self.cnn(protein_seq_enc)
prot_enc, prot_attention = self.protein_attention_layer(protein_seq_enc, protein_seq_enc, protein_seq_enc)
seq_enc, sequence_attention = self.sequence_attention_layer(sequence_enc, sequence_enc, sequence_enc)
prot_enc, seq_enc, prot_seq_attention, seq_prot_attention = self.reciprocal_attention_layer(prot_enc,
seq_enc,
seq_enc,
prot_enc)
prot_enc = self.ffn_protein(prot_enc)
seq_enc = self.ffn_seq(seq_enc)
return prot_enc, seq_enc, prot_attention, sequence_attention, prot_seq_attention, seq_prot_attention
class ReciprocalLayer(nn.Module):
def __init__(self, d_model, d_inner, n_head, d_k, d_v):
super().__init__()
self.sequence_attention_layer = MultiHeadAttentionSequence(n_head, d_model,
d_k, d_v)
self.protein_attention_layer = MultiHeadAttentionSequence(n_head, d_model,
d_k, d_v)
self.reciprocal_attention_layer = MultiHeadAttentionReciprocal(n_head, d_model,
d_k, d_v)
self.ffn_seq = FFN(d_model, d_inner)
self.ffn_protein = FFN(d_model, d_inner)
def forward(self, sequence_enc, protein_seq_enc):
prot_enc, prot_attention = self.protein_attention_layer(protein_seq_enc, protein_seq_enc, protein_seq_enc)
seq_enc, sequence_attention = self.sequence_attention_layer(sequence_enc, sequence_enc, sequence_enc)
prot_enc, seq_enc, prot_seq_attention, seq_prot_attention = self.reciprocal_attention_layer(prot_enc,
seq_enc,
seq_enc,
prot_enc)
prot_enc = self.ffn_protein(prot_enc)
seq_enc = self.ffn_seq(seq_enc)
return prot_enc, seq_enc, prot_attention, sequence_attention, prot_seq_attention, seq_prot_attention
|